Warehousing trends and innovations

Prediction is very difficult, especially if it’s about the future. Niels Bohr

Learning objectives 📘

• Discuss warehouse trends and innovations.
• Identify the challenges of smart warehouses.

Niels Bohr said “Prediction is very difficult, especially if it’s about the future”. We are now at a stage where the challenges within the supply chain are necessitating a change of mindset amongst warehouse operators, with automation and robotics being considered a viable and at times necessary alternative to large amounts of labour. This is now being seen globally.

We are also seeing companies such as Amazon introducing robotics on a large scale. Amazon recently received a patent for its idea of a warehouse in the sky with orders being fulfilled in an airship and the items being delivered by drones. Farfetched it may seem at the moment but these companies are pushing the envelope in order to meet ever increasing and complicating challenges. The development of Artificial Intelligence (AI) and Virtual Reality (VR) will also have a significant effect on warehousing in the future.

— Warehouse Management: A Complete Guide to Improving Efficiency and Minimizing Costs in the Modern Warehouse (Richards 2018, 646).

External change forces

Figure 1: The confluence of external change forces in the 21st century has dramatically changed the economic landscape and provided an opportune business climate for the development of global supply chains and supply chain management (Coyle et al. 2020).

“Six major external forces are driving the rate of change: globalization, technology, organizational consolidation, the empowered consumer, sustainability, and government policy and regulation. The confluence of these factors in the twenty-first century has dramatically changed the economic landscape and provided an opportune business climate for the development of global supply chains and supply chain management.”

— Supply Chain Management: A Logistics Perspective (Coyle et al. 2020, 7).

Challenges

According to Richards (2018), the greatest challenge is the growth in e-commerce, resulting in:

• smaller, more frequent orders;
• shorter order lead times;
• next-day delivery will be the norm;
• same-day delivery in certain circumstances;
• an increase in home delivery;
• increased personalization of items.

Other challenges include:

• A greater proliferation of product lines or SKUs.
• A requirement for accurate information in real time.
• The need to achieve the perfect order to ensure competitiveness.

The continuing challenges faced by companies are many and varied.

The following list is not exhaustive but details the number of challenges being faced by companies today and those likely to be faced in the future.

One of the biggest challenges is the growth in e-commerce. This is already resulting in the following:

• Smaller, more frequent orders.
• Shorter order lead times.
• Next day delivery will be the norm.
• Same-day delivery in certain circumstances.
• Increase in home delivery.
• Increase in personalization of items.

Other challenges are as follows:

• A greater proliferation of product lines or SKU.
• A requirement for accurate information in real time.
• The need to achieve the perfect order to ensure competitiveness.
• A requirement for lower overall inventory and overall cost reduction.
• An increase in the cost of land.
• Increasing labour costs.
• Ageing populations in many countries.
• A shortage of skilled labour.
• A shortage of management knowledge and expertise.
• The moral question as to what to do with displaced staff on a large scale.
• Security of products.
• Traceability of items.
• Cold chain storage and distribution.
• A greater need for systems integration–ERP systems with WMS and TMS for example.
• The growth in cloud-based systems and the reluctance in some quarters to accept them–especially in terms of data security.
• Overall data security.
• Local and global regulations and protectionism.
• Sustainability demands–pressure for more environmentally friendly warehouses.
• More buzz words such as Industry 4.0, M to M (Machine to Machine) and the Internet of Things!

Suppliers, users and academics are having to look at ways of overcoming these challenges. New technology is continually being introduced and, as Mark Strand says, “the future is now” in many respects. Concepts such as fully automated warehouses, carbon-neutral buildings, hybrid trucks, robotics, drones, voice and optically guided warehouse operations are all in their various stages of development.

— Warehouse Management: A Complete Guide to Improving Efficiency and Minimizing Costs in the Modern Warehouse (Richards 2018, 646–47).

Factors shaping the future of warehousing

The following factors are supported by the cited literature and primary records:

• Economic and population growth: Increasing demand for goods and services will drive the need for more efficient and scalable warehousing solutions.
• Sustainability: Warehouses will need to adopt greener practices to reduce their environmental impact and meet regulatory and consumer expectations.
• Pressure to collaborate: Shared resources and collaborative logistics will become essential to optimize costs and improve efficiency.
• Shortage of warehouse space: Innovative solutions like underground or vertical warehouses will address the growing scarcity of land.
• Generation Z: This tech-savvy generation will push for digitalization and hyperautomation in warehouse operations.
• 3D printing: On-demand manufacturing will reduce inventory needs and transform traditional warehousing models.

The factors on this slide come from two evidence layers. Richards supplies the long-run warehouse context. Public primary records support the named company, facility, platform, and hyperautomation examples. Citations are attached to the specific claim they support; Richards is not a blanket source for the composite.

Economic and population change

Globally we have a growing but ageing population. This means that land will be at a premium but also there are likely to be labour shortages in key areas which suggests a potential growth in automation but also the need for elder friendly workplaces and in some countries people will be required to prolong their working lives.

The growing economies of not only the BRICS countries (Brazil, Russia, India, China and South Africa) but also those of the developing world where consumerism and the growth of the internet will put even greater pressure on consumer product manufacturers and their warehouses. Consumer awareness and demand for new products and services will continue to increase.

— Warehouse Management (Richards 2018, 455–56).

The demographic argument combines demand growth, land pressure, labor scarcity, and greater automation. It does not say that automation is always feasible or that older workers should be displaced; it highlights the need for both technology skills and accessible workplaces.

Sustainability and collaboration

Sustainability will play a significant role in supply chain operations in the future. The green lobby will look to the supply chain for initiatives in terms of alternative energy use, reduction in CO2 emissions, reduction in waste and water usage and the use of alternative forms of transport. This will include intermodal transport initiatives as well as fuel-efficient MHE. Consumers and retailers will also be encouraged to source locally leading to an increase in demand for neighbourhood warehouses.

Fuel and energy costs will continue to rise as fossil fuels decline. Companies will look to warehouse automation and the use of greener vehicles whilst developers and warehouse operators will be encouraged to consider solar panels, wind turbines and the use of waste product for energy production.

An increasing pressure on companies to collaborate and share resources. Many warehouses and, for that matter, transport modes are under-utilized, so pressure from the green lobby and also continued pressure to further reduce costs will encourage more companies to collaborate.

— Warehouse Management (Richards 2018, 455–58).

XPO and Nestle provide a concrete collaboration case. Their official release describes a 638,000-square-foot UK distribution facility developed with Swisslog, using robotics, automated sorting, and XPO analytics. It also states, “The site includes an innovation hub that also functions as an incubator.” The hub supports real-world logistics trials; this claim comes from the company release, not a Richards figure (XPO Logistics, Inc. and Nestle 2021, opening description and innovation-hub paragraphs).

Scarce space: vertical, underground, and flexible capacity

The majority of warehouses are likely to be high bay (at least 12-metre eaves according to the BNP Paribas report) with some lower height warehouses or sections of warehouses to act as cross docking operations and areas for value-adding services. The continued growth in third-party logistics and shared-user warehousing is also likely to see larger warehouses with the potential for sub-division.

The increase in product lines could also see greater use of mezzanine floors.

— Warehouse Management (Richards 2018, 461–63).

Vertical space is one response; reusing subterranean space is another. CMS documents limestone-cave warehousing in Kansas City and archive storage in a Cheshire salt mine. For Heathrow it records an approved proposal, not a completed facility: “Plans have been approved to create a 175,000 square metre underground logistics space near to Heathrow Airport” (CMS 2018, 25, sec. “Subterranean”).

Fabric’s primary announcement describes an operating micro-fulfillment site under Shalom Meir Tower in Tel Aviv. The triangular, low-ceiling facility is about 18,000 square feet and supports one-hour and same-day grocery orders. Fabric states, “It is a full-service grocery site with three temperature zones,” covering fresh, ambient, chilled, and frozen products (Fabric 2019, opening facility description).

Flexible capacity addresses a different form of scarcity. JLL reports growing occupier demand for shorter-term logistics space, with digital platforms matching available capacity to users and warehousing-as-a-service supplementing or replacing a fixed network. Its summary is: “We are seeing increasing occupier requirements for flexible logistics space available ‘on demand’” (Cross and Duncan 2024, sec. “Talking point: ‘Flexible Logistics’”).

The platform counts must be time-stamped. TIMOCOM reported in 2019 that its warehouse exchange listed “meer dan 7.000 aanbieders van magazijnruimte in 44 Europese landen”; this is a historical count, not a current total (TIMOCOM GmbH 2019, sec. “Waarom een opslagbeurs gebruiken?”). Wincanton’s 2020 annual report described oneVASTwarehouse as “A cloud-based virtual warehouse” that brought buyers and sellers together online, reporting 46.2 million square feet across more than 340 sites (Wincanton plc 2020, 7).

Generation Z and hyperautomation

According to Kewill (2013) Generation Z—the post-90s generation—has entered the market, as both consumers and employees. Having grown up using e-mail, social networking and communications technology such as mobile/smart phones, MP3 players, laptops/tablets and games consoles including virtual reality, they have never known a world without them. Adept at switching between multiple platforms, formats and devices, they expect to utilize the technology they are familiar with in the work environment, accelerating the prevalence of Bring Your Own Device (BYOD) in industries heavily reliant on the timely transfer of data, such as logistics.

— Warehouse Management (Richards 2018, 455–56).

Gartner’s 2021 release reports that supply-chain leaders expected digitally native Gen Z entrants to accelerate digitalization and supply-chain autonomy. It defines hyperautomation as combining technologies including robotic process automation, machine learning, artificial intelligence, and other tools (Gartner 2021, opening findings and the Pierfrancesco Manenti paragraph). This supports the digitalization and hyperautomation claim without assigning the Gartner statement to Richards.

3D printing

One other piece of technology I want to share with you is 3D printing or additive manufacturing. 3D printing was originally developed as an automated method of producing prototypes. Although there are several competing technologies, most work on the basis of building up layers of material (sometimes plastic, ceramics or even metal powders) using a computer-aided design. Hence, it is referred to as an ‘additive’ process; each layer is ‘printed’ until a three-dimensional product is created.

A report by Transport Intelligence suggests that 3D printing ‘is already very good at producing products (even with moving parts) which previously would have required the assembly of multiple components,’ and that by ‘eliminating the assembly phase there will be huge savings for the manufacturer in terms of labor costs.’ 3D printing-based production can also reduce or eliminate storage, handling and distribution costs.

— Warehouse Management (Richards 2018, 463–64).

The warehouse implication is conditional. Distributed printing can reduce some finished-goods and service-parts inventory while creating demand for raw materials, digital design control, quality assurance, and new replenishment flows. It changes what is stored and where; it does not eliminate logistics.

Smart warehouse

A smart warehouse is an automated, unmanned, and paperless facility that streamlines the operations of pickup, delivery, and bookkeeping. (Liu et al. 2018)

The fundamental characteristics of smart warehouses can be categorized as follows (Zhen and Li 2022):

• Information Interconnection: Seamless integration of the virtual and physical worlds through cyber-physical systems (CPSs).
• Equipment Automation: Enhanced productivity and reduced reliance on manual labor through advanced automation technologies.
• Process Integration: Streamlined coordination and management of warehouse operations.
• Environmental Sustainability: Focus on minimizing energy consumption and reducing carbon emissions.

“A smart warehouse is an automated, unmanned, and paperless warehouse when conducting the operations of pickup, delivery, and bookkeeping. In this paper, we focus on the technologies towards building smart warehouses [1]. In a traditional warehouse, the operations of pickup, delivery, and bookkeeping are accomplished by storekeepers. There are several drawbacks of conventional warehouses. First, it is time-consuming to store/fetch inventories into/from them. Second, the usage of storekeepers is a waste of human resources. Finally, it is not environmentally friendly to record the stocks using account books. Therefore, it is profoundly essential to make traditional warehouse smart.”

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, 1).

“Driven by industry needs, academic research on smart warehouses has been a hot topic. Some scholars have put forward their idea on smart warehouses and smart logistics. According to the summary of McFarlane et al. (2016), the term”intelligent logistics” refers to different logistics operations, which are planned, managed, or controlled in a more intelligent way compared with conventional solutions. Wen et al. (2018)’s work shows a strong requirement for automation and intelligence technology in smart logistics. In our opinion, smart warehouses are the collection of smart technologies in warehouse section and a series of operations management practices to make warehouses operate in a “smarter” way. Smart warehouses have developed into an integral of cutting-edge technologies, warehouse processes, and warehouse operations management. The basic characteristics of smart warehouses can be classified into the following perspectives:

Information interconnection. Information interconnection involves the top-level design of smart warehouses. It is the base of smart warehouses and operational management. Based on technology derived from the Internet of Things (IoT), cyber–physical systems (CPS), and other emerging technology, information flow can be shared and processed by numerous logistics nodes and thus produce extra values.

Equipment automation. Equipment automation describes the characteristics of smart warehouse at strategic and tactical levels. Automation is the technical support of the smart warehouse. Equipped with automatic facilities, smart warehouses can achieve high automation levels in warehouse activities. Equipment automation can improve warehouse productivity while reducing the need for manual labor. Besides, the operations management of smart warehouses pays more attention to equipment characteristics on strategic-level decisions and product characteristics on tactical-level decisions, providing a holistic view of the technology, and improves the precision of decision-making.

Process integration. Process integration is the requirement of smart warehouse operations management and functions as operational support in the framework. Process integration tries to implement overall planning among various warehouse processes and focuses on the new operational problems arising in the operation of smart warehouses. The objective of process integration is to achieve coordination while eliminating discordance of warehouse operations management.

Environmental sustainability. Environmental sustainability is the future of smart warehouses, supported by equipment automation and process integration. The sustainable development of smart warehouses concerns the problems relevant to the environment, such as energy consumption and carbon emission. The operations management of smart warehouses on strategic, tactical, and operational levels should implement in an eco-friendly way to create a sustainable roadmap in the warehouse section.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 32).

• Smart Warehouses of the Future | Delivered | Global (dhl.com)

Information interconnection

In the last decades, CPS and IoT, which emerged as the extension of computer networks into every device and object, have become essential supporters for enabling the interconnection of logistics activities. CPS and IoT turn the ideal concept into a reality where goods, logistics facilities, and logistics elements are connected and communicated (Ben-Daya et al., 2017; Lee et al., 2018; Yang et al., 2019). In today’s warehouse practice, RFID, warehouse management systems (WMS), augmented reality (AR), reinforced learning, and other emerging technologies enhance the interconnection between different processes and entities (Dou et al., 2015; Bottani and Vignali, 2019; Sartoretti et al., 2019). Interconnection technology provides ways to establish warehouse information collection and exchange, which could be a leading factor for future warehouse development. IoT and CPS technologies, including radio frequency, pick-to-light, and pick-by-voice, are commonly used in the warehouse picking process.

The brief introduction to interconnection technology could bring some insights for a better understanding of the smart warehouse. To formulate a way of choosing suitable interconnection technology, Hassan et al. (2015) identify the affecting factors of warehouse automatic identification technology through a literature review. The major affecting factors for selecting automatic identification technology include organizational, operational, structural, resources, external, environmental, and technological factors.

RFID is a technology for automatic object identification and data collection based on radio waves. Among the interconnection technologies, RFID is considered a choice for positioning, identifying, information interaction, and warehouse management in the smart warehouse. Its strengths include non-contact sensing ability, high flexibility, configurability, high efficiency, and low cost (Tao et al., 2014). Giusti et al. (2019) work on a generalized performance comparison model of different RFID implementations. The benefit related to informatization on warehouse performance has been investigated. A probabilistic risk assessment model is applied to identify and quantify the delay value in transfer time.

The application of RFID in smart warehouses mainly focuses on inventory tracking, localization, and warehouse management. Information exchange, which is the main function of the RFID system, significantly impacts the warehouses’ efficient communication. Information accuracy is essential for managing warehouse information exchange efficiency and effectiveness (Zhong et al., 2015). Object localization based on RFID technology serves as a major function of RFID warehouse management. One of the most challenging issues in RFID-based applications is the acquisition of location information of either the tags or readers. Thus, RFID-based localization (especially three-dimensional localization) has been an active research topic (Lu et al., 2018).

The use of tracking inventory is also investigated by researchers (Shahzad and Liu, 2015; Zhou et al., 2017; Mo and Li, 2019; Gareis et al., 2021). Missing tags searching is another RFID object location and inventory tracking application (Chen et al., 2017). IoT and CPS technologies can also influence the way of warehouse design and operation. RFID tags are attached to products at the item level instead of the pallet level, providing a way to support warehouse storage assignment with higher precision (Choy et al., 2017). RFID can provide infrastructure for collaborative warehouse order fulfillment of decentralized management to improve system reaction capabilities (Reaidy et al., 2015). The real-time information exchange and collection of RFID can provide a decision support tool for order fulfillment (Lam et al., 2015). Furthermore, the energy-saving improvement of RFID could be a new point for the warehouse system.

In some research about conflict-free routing in the robotized warehouse, CPS is used for information exchange and pre-detection of potential collisions. Collisions can be detected by comparing the occupancy of the time window, and some corresponding collision avoidance strategies can be developed for conflict-free routing (Lee et al., 2019; Keung et al., 2020).

WMS is a computerized database application widely used by the logistics section for warehouse control and optimization. A common WMS includes warehouse processes such as receiving, storage, order picking, packing, and shipping. IoT provides ways for information collection and exchange for WMS.

The implementation of IoT-based WMS can also facilitate warehouse operation. Qiu et al. (2015) introduce a framework of IoT-enabled Supply Hub in Industrial Park (SHIP). The innovations of IoT-enabled SHIP include different information sharing levels, information structure, and decision-making tools. Lee et al. (2018) propose an IoT-based WMS integrated with the fuzzy logic technique. Based on the case data, the IoT-based WMS shows its competitiveness in improving warehouse productivity, picking accuracy, efficiency, and robustness in order variability.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 38).

Information interconnection

• A CPS¹-based smart warehouse contains four main components:
  • CPS devices
    • Radio-frequency identification (RFID), narrowband IoT (NB-IoT), Wi-Fi, near-field communication (NFC), fifth-generation mobile networks (5G), global positioning systems (GPS), wireless sensor networks (WSNs), and robotics.
  • Inventories
    • Status and location context according to the data reported by the attached devices.
  • Robots
    • Collaboratively perform tasks that are repetitive or harmful to people.
  • Human beings
    • Use recognized gestures or other activities to perform warehouse operations.
• Information interconnection technology provides ways to establish warehouse information collection and exchange.
• The implementation of IoT-based WMS can also facilitate warehouse operation.

In the last decades, CPS and IoT, which emerged as the extension of computer networks into every device and object, have become essential supporters for enabling the interconnection of logistics activities. CPS and IoT turn the ideal concept into a reality that goods, logistics facilities, and logistics elements are connected and communicated (Ben-Daya et al., 2017; Lee et al., 2018; Yang et al., 2019). In today’s warehouse practice, RFID, warehouse management system (WMS), augmented reality (AR), reinforced learning, and other emerging technologies enhance the interconnection between different processes and entities (Dou et al., 2015; Bottani and Vignali, 2019; Sartoretti et al., 2019).

Interconnection technology provides ways to establish warehouse information collection and exchange, which could be a leading factor for future warehouse development.

IoT and CPS technologies, including Radio frequency, pick to light, and pick by voice, are commonly used in the warehouse picking process.

The brief introduction on interconnection technology could bring some insights for a better understanding of the smart warehouse. To formulate a way of choosing suited interconnection technology, Hassan et al. (2015) identify the

affecting factors of warehouse automatic identification technology through a literature review. The major affecting factors for selecting automatic identification technology include organizational, operational, structural, resources, external, environmental, and technological factors.

RFID is a technology for automatic object identification and data collection based on radio waves. Among the interconnection technologies,

RFID is considered a choice for: Inventory tracking, localization, and warehouse management.

• positioning,
• identifying,
• information interaction, and
• warehouse management in the smart warehouse. Its strengths include non-contact sensing ability, high flexibility, configurability, high efficiency, and low cost (Tao et al., 2014). Giusti et al. (2019) work on a generalized performance comparison model of different RFID implementations. The benefit related to informatization on warehouse performance has been investigated. A probabilistic risk assessment model is applied to identify and quantify the delay value in transfer time.

Information exchange, which is the main function of the RFID system, significantly impacts the warehouses’ efficient communication. Information accuracy is essential for managing warehouse information exchange efficiency and effectiveness (Zhong et al., 2015). Object localization based on RFID technology serves as a major function of RFID warehouse management. One of the most challenging issues in RFID-based applications is the acquisition of location information of either the tags or readers. Thus, RFID-based localization (especially threedimensional localization) has been an active research topic (Lu et al., 2018). The use of tracking inventory is also investigated by researchers (Shahzad and Liu, 2015; Zhou et al., 2017; Mo and Li, 2019; Gareis et al., 2021). Missing tags searching is another RFID object location and inventory tracking application (Chen et al., 2017).

IoT and CPS technologies can also influence the way of warehouse design and operation.

RFID tags are attached to products at the item level instead of the pallet level, providing a way to support warehouse storage assignment with higher precision (Choy et al., 2017). RFID can provide infrastructure for collaborative warehouse order fulfillment of decentralized management to improve system reaction capabilities (Reaidy et al., 2015). The real-time information exchange and collection of RFID can provide a decision support tool for order fulfillment (Lam et al., 2015).

Furthermore, the energy-saving improvement of RFID could be a new point for the warehouse system.

In some research about conflict-free routing in the robotized warehouse,

Cyber–physical systems (CPS):

• Information exchange and pre-detection of potential collisions (conflict-free routing in robotized warehouses).

Internet of Things (IoT):

• Provides ways for information collection and exchange for WMS.

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 38).

1. CPS: cyber-physical system

Efficient communication scheduling

How to guarantee efficient communication when collecting sensing data from thousands of CPS devices?

• RFID is more suitable for localization in Industry 4.0:
  • GPS (works well only outdoors)
  • Wi-Fi (locates a person carrying a cellphone)
  • Bluetooth (limited battery)
• Future: tools and products affixed with sensors and RFID tags.
• Benefits of RFID:
  • No line-of-sight requirement (\(n\) objects can be identified simultaneously).
  • Battery-free operation (passive tags have a nearly unlimited service life).
  • Low cost (five cents).
  • Small size (as thin as paper).

Efficient Communication Scheduling

In the future smart warehouse, each screw, tool, or product may be affixed with a sensor or RFID tag for inventory. Due to the contradiction between the large number of sensing devices and the limited communication resources, it is non-trivial to investigate efficient communication scheduling strategies when collecting sensing data from large-scale CPS devices. By saying large-scale, it means thousands of CPS devices working together. For example, in a smart warehouse, there can be tens of thousands of CPS devices such as RFID tags and readers, BLE beacons, Wi-Fi APs, robots, etc. Compared with the existing automatic identification techniques such as barcode, sensor, and Bluetooth, RFID has promising advantages of no requirements on line-of-sight (multiple objects can be identified simultaneously), battery-free (passive tags have almost infinite service time), and small size (as thin as paper), which make RFID more suitable for large-scale deployment in future smart industries.

Therefore, we discuss RFID communication scheduling in this paper. A general solution is to identify all tags’ IDs for inventory. There are two types of tag identification protocols: Aloha-based protocols and tree-based protocols.

• Aloha-based protocols: These are a type of Time Division Multiple Access (TDMA) mechanism. Multiple tags contend for multiple time slots, and a tag can be successfully identified in a slot when only one tag responds in the selected slot.
• Tree-based protocols: In these protocols, the RFID reader broadcasts a 0/1 string to query the tags. A tag will respond with its ID once it finds that the queried string is the prefix of its ID. If multiple tags respond, the reader attaches a bit of 0/1 to the end of the previous string. A longer queried string entails a smaller probability of tag collisions. A reader identifies a tag ID when only one tag responds.

Although we can gather the most comprehensive data via tag identification protocols, it leads to much overhead regarding time efficiency or energy efficiency. A general solution may not be optimal for a specific problem, e.g., missing tag detection/identification, unknown tag detection/identification, RFID tag cardinality estimation, etc. This is because we only want to know which tags are missing or unknown, and the tag cardinality instead of the exact tag IDs. Hence, plenty of attention has been paid to the detection/identification of missing/unknown tags and the tag cardinality estimation.

• Missing tags: These are tags that should be present in the warehouse but are absent now, which may be caused by theft.
• Unknown tags: These are tags that should not be in the warehouse but are present now, which may be caused by misplacement.

Detection is to judge whether there is an anomaly tag or not; identification is to discover all the anomaly tags accurately. The literature [4] has conducted a comprehensive survey of the existing anomaly tag (missing [5,6] or unknown [7–9]) pinpointing protocols, particularly emphasizing the evolution road of the related protocol development. After that, the authors also conducted a taxonomy of the existing protocols and summarized their pros and cons, respectively. Hence, this paper will not discuss the details of the protocols that deal with missing/unknown tags again.

In this paper, we pay more attention to the existing RFID cardinality estimation protocols. Although RFID identification protocols can be used to obtain the exact tag IDs and obviously get the tag cardinality, the involved execution time is extremely high, particularly when the number of tags is large (proportional to the number of tags). In scenarios when we just want to know the approximate number of tags instead of exact tag IDs, it is not efficient to execute the tag identification protocols.

Hence, a great deal of effort has also been made by academic communities to study the tag cardinality estimation problem, which is defined as follows:

Given a set of RFID tags whose cardinality is n, a confidence interval α, and a reliability β, we want to obtain the estimation result n̂ such that Pr{|n̂ − n| < n × α} > β.

Kodialam et al. took the first step to study this problem and proposed the cardinality estimation schemes called Unified Simple Estimator (USE) and Unified Probabilistic Estimation (UPE), which use the number of empty or collision slots to estimate population sizes [10].

• USE and UPE: These are used for RFID systems that use Aloha-based protocols as the MAC (media access control) layer communication mechanisms. The basic working principle of these two protocols is as follows:
  • The reader announces a frame size for all tags and then initializes a time-slotted frame.
  • All the tags randomly select time slots in the forthcoming time frame to respond.
  • The reader records the number of empty slots (no tag response), singleton slots (just one tag response), and collision slots (two or more responses).

Intuitively, the more tags there are in the vicinity of the reader, the reader observes more collision slots and fewer empty slots. Moreover, the authors built a functional relationship between the number of tags and the number of empty/collision slots. Then, the number of observed empty/collision slots can be used to calculate the number of tags. The estimation result from one-time frame may not be accurate due to estimation deviation. Hence, multiple frames of estimation are repeated to achieve a fine-grained estimation result.

Currently, there are a bunch of RFID estimation protocols. Their performance differs because they use the observed slot status differently.

• ART (Average Run-based Tag estimation): Shahzad et al. used the average run length of non-empty slots (i.e., the average length of continuous non-empty slots) for cardinality estimation. ART is faster for obtaining the fine-grained estimation result because the used estimator has smaller variance.
• MLE (Maximum Likelihood Estimator): Li et al. revisited the problem of cardinality estimation from the new view of energy efficiency and proposed MLE.

The above RFID estimation protocols are all designed for RFID systems that run the Aloha-based MAC communication mechanisms.

• PET (Probabilistic Estimation Tree): Zheng et al. studied this problem for RFID systems that run tree-based communication mechanisms and proposed PET protocols.

The above estimation protocols can only estimate the number of tags that are currently present in the systems. However, some RFID systems are dynamic because some tags are moved out of the system, and some new tags are added to the system. Moreover, the above methods cannot tell how many tags are absent and how many tags are new.

For such dynamic scenarios:

• Xiao et al. compared the observed frame status and the expected frame status and used the slot status changes to estimate the number of absent/remaining/new tags [14].
• Liu et al. found that the estimation protocol in [14] requires the reader to monitor the whole long time frame even if the time frame is quite long. To this end, they proposed the Sampling-based Key Tag Tracking (S-KT) protocol, in which an RFID reader only needs to observe the status change of expected singleton slots corresponding to key tags instead of the whole time frame.

In fact, the ratio of key tags to all current tags is small because key members are usually rare. As a result, even when the whole time frame is long, the number of expected singleton slots could still be small.

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, 3–5).

Accurate and robust localization

• RFID localization systems are evaluated against four metrics:
  • Availability:
    • System consists of commercial off-the-shelf (COTS) devices.
    • Localization algorithms are only installed as software modules.
  • Mobility:
    • System covers a vast region in a mobile manner to save costs.
  • Reliability:
    • System provides fine-grained localization accuracy (i.e., cm-level) with high confidence.
  • Scalability:
    • System is scalable to a large number of tags.
    • Depends on computational complexity¹ of localization algorithms.

“Accurate and Robust Localization

Accurate object localization benefits various context-aware applications in the era of Industry 4.0, e.g., automatic product tracking and sorting. Currently, there are various localization techniques such as GPS, Wi-Fi, Bluetooth, and RFID. For large-scale warehouse in future Industry 4.0, RFID has attractive advantages over the other techniques. GPS can work well for the outdoor environment (Google Maps, driving navigation) but does not work for the indoor environment. Wi-Fi can pinpoint the location of the human that takes a smartphone. However, it will be nearly impossible to affix each product with a phone. Bluetooth requires each product has a battery-powered beacon device. The limited battery volume severely shortens the lifetime of the localization services. Some other technologies are rising for localization and human activity recognition such as 5G. However, 5G is still on its way and cannot be accessed ubiquitously. Therefore, the relating technology for localization and human activity recognition are not well developed. Compared with the above techniques, passive RFID tags are low-cost (five cents for each tag), small-size (as thin as a paper), and battery-free (almost infinite lifetime), which make RFID more suitable for localization in Industry 4.0. The existing RFID localization approaches can be classified into two types: RSS-based approaches (received signal strength), and phase-based approaches. In the infancy of the RFID localization study, Ni et al. exploited the RSS value of RFID tags to implement a localization system called LANDMARC [19]. The basic working principle of LANDMARC is as follows. The manager of a warehouse deploys many reference tags in the surveillance region in prior. As a matter of fact, RSS value closely depends on the distance between tag and reader antenna. Thus, RSS values of the nearby tags should also be similar. The RFID readers keep interrogating the concerned tags as well as the reference tags. For each tag reading, we can obtain not only the tag IDs but also the RSS. Then, we find out k reference tags whose RSS values are the closest to that of the concerned tag and use the weighted average of the locations of these k reference tags as the location of the target tag. However, RSS is an unreliable location indicator, especially for UHF tags [20]. Recently, the researchers shifted attention to use RF phase information when addressing the problem of RFID localization. The Angle of Arrival (AOA) based methods make use of multiple reader antenna arrays [21]. For one antenna array, the phase difference is used to measure the angle between the target tag and this antenna array plane. Using two antenna arrays, we can draw two lines based on the measured angles, and the intersection point is expected to be the tag location. Wang et al. implemented a localization system called PinIt [22], which is based on comparing the multi-path effects instead of RSS. Specifically, they use a moving antenna to measure the multi-path profiles of the reference tags at known positions in advance. Then, PinIt locates the target tag according to the insight that nearby tags have similar multi-path effects. For the anonymous RFID system, PinIt is hard to use because we know nothing about the reference tags. Yang et al. proposed the location tracking system called Differential Augmented Hologram (DAH) [23], which exploits the target tag’s mobility to construct a virtual antenna array by using readings from a few physical antennas over time. They further proposed a comprehensive solution to accurately recover the tag’s moving trajectories and its locations, relaxing the assumption of knowing tag’s track function in advance. DAH partitions the monitoring region into grids, and use the collected phase profile to calculate the probability that the target tag locates in each grid. The grid with the highest probability is treated as the location of the target tag. The experimental results reveal that DAH can achieve cm-level accuracy. However, DAH requires precise calibration of the reader antennas. More seriously, for the large surveillance region, DAH needs to deploy multiple sets of reader antennas, which costs a lot. However, RF-Scanner begets high overhead regarding computational complexity, and thus cannot satisfy the real-time requirement for tag-dense RFID systems. To evaluate the existing RFID localization systems, we need to concern with four metrics. Availability: whether the localization system only consists of the Commercial Off-The-Shelf (COTS) devices, and the proposed localization algorithms are only installed as software modules. Mobility: whether a single localization system could be able to cover a vast region in a mobile manner to save cost Reliability: whether the system could provide fine-grained localization accuracy (i.e., cm-level) with high confidence. Scalability: whether the computational complexity of the RFID localization algorithms is quite low so that it is scalable to a large number of tags. In Table 1, we have summarized the pros and cons of the existing RFID localization systems. In terms computational complexity, we have to explain the involved notations. x represents the number of tag replies received by the reader; ξ is the number of reference reader antennas; and γ is a factor that is determined by localization granularity. From this table, we can easily find that the DAH and RF-Scanner can provide the most accurate localization results. Compared with DAH, RF-Scanner is more suitable for the large-scale environment.”

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, 6).

1. Computational complexity may depend on number of tag replies received by the reader, number of reference reader antennas, localization granularity.

Multi-robot collaboration

• Multi-robot tasks involve:
  • Multiple jobs executed on different robots.
  • End-to-end deadline requirements.
  • Local coordination among robots.

How to integrate global task scheduling and local coordination in MRS¹?

“Multi-Robot Collaboration No matter what the multi-robot applications are, one of the fundamental problems is to integrate global task scheduling and local coordination. Consider a multi-robot application and a user-defined task assigned to it using some programming model or middleware [26]. The task may consist of multiple subtasks, and the technique of task scheduling is needed. After the subtasks are assigned to each robot, the robots are supposed to execute the subtasks with local coordination. In the following subsections, we introduce the techniques towards solving the problems of task scheduling and local coordination in multi-robot system, respectively.

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, 7).

1. MRS: Multi-robot systems.

Multi-Robot task scheduling vs. local coordination

Taxonomy of MR task scheduling

Taxonomy of multi-robot collaboration

Figure 2: To integrate global task scheduling and local coordination in MRS, we need to consider both aspects simultaneously.

Multi-Robot task scheduling

Figure 3: Taxonomy of MR task scheduling (Liu et al. 2018)

• Scheduling
  • Centralized:
    • A central platform decomposes the task into multiple subtasks and allocates them to the robots.
    • Robots execute the assigned subtasks.
  • Fully-distributed:
    • Robots negotiate with each other and play equal roles in task scheduling.
  • Semi-distributed:
    • The MRS is divided into several robot clusters.
    • Each cluster schedules its tasks independently.
    • A central platform assigns tasks to the clusters.
• Criticality
  • Soft real-time (SRT):
    • Tasks are performed in a best-effort manner (e.g., missing deadlines are tolerated).
  • Hard real-time (HRT):
    • Safety-critical applications (e.g., human-assisted robotic manipulation) where meeting deadlines is essential.
• Technique
  • Market-based:
    • Tasks are divided into a set of subtasks.
    • Robots bid to carry out these subtasks.
    • Robots negotiate with each other to maximize their self-interests.
  • Bio-inspired:
    • Observed rules from nature are applied to solve scheduling problems.
    • Example: Mimicking bird flocking behavior for multi-robot pattern formation.
  • Hybrid:
    • Common in HRT task scheduling.
    • Combines design-time computations with runtime allocation to satisfy timing constraints.

Source section: 4.1. Multi-Robot Task Scheduling

Many multi-robot tasks involve multiple jobs executed on different robots with specified end-to-end deadline requirements. These tasks may be deployed to an MRS at any time and on any robot during run-time. For example, a new package delivery task may be deployed on an MRS while a persistent monitoring task is in process. As a result, resource competitions among these tasks can happen on any robot and is unpredictable. Furthermore, an individual robot often does not have full information about the working status of the other robots since the robots form a distributed system. Therefore, scheduling decisions made by individual robots about their local job execution orders may not be optimal for the tasks to which the jobs belong concerning meeting the tasks’ end-to-end deadlines. Task scheduling in distributed systems has been proven to be NP-hard (non-deterministic polynomial-time hard) [27], and the research community has been developing various heuristic strategies to schedule tasks for MRS. We categorize the existing approaches according to three metrics, namely criticality, resource management, and technique as shown in Figure 2. The dimension of “scheduler” represents the member who carries out scheduling. The scheduling can be performed in a centralized, semi-distributed, or fully-distributed way. The scheduling is performed in a “centralized” fashion if a central platform decomposes the task into multiple subtasks and allocate the subtasks to the robots. In this approach, the robots are only responsible for receiving the result subtasks from the central platform and performs them. “Fully-distributed” scheduler means the robots themselves serve as the scheduler. In this fashion, all the robots negotiate with each other and play equal roles. The “semi-distributed” scheduling approach combines the centralized and fully-distributed approaches. The whole MRS is divided into several clusters. Each cluster of robots schedule its tasks independently, and a central platform is responsible for assigning tasks to the clusters. Concerning criticality, different multi-robot applications have a different degree of criticality. According to the criticality of the applications, multi-robot task scheduling can be divided into soft real-time (SRT) task scheduling and hard real-time (HRT) task scheduling. For those safety-critical applications, e.g., human-assisted robotic manipulation and operating medical equipment, HRT is required since human beings are involved, and safety issues pop up. For other applications, such as package delivery, missing deadlines can be tolerated, and soft real-time (SRT) is preferred. SRT task scheduling is performed in a best-effort way. There are various techniques for multi-robot task scheduling such as market-based [28,29], bio-inspired [30,31], and hybrid approaches [32]. In the market-based approach, the task is divided into a set of subtasks and the robots bid to carry out these subtasks. Since the robots are assigned limited amounts of money, they have to negotiate with each other to maximize their self-interests. The market-based approach is widely used for multi-robot searching [33] and patrolling [34]. Bio-inspired task scheduling approaches apply the observed rules from nature to solving scheduling problems. For example, mimicking the birds flock for multi-robot pattern formation has been a success story [35]. Generally speaking, market-based and bio-inspired approaches are used for SRT multi-robot task scheduling. When it comes to HRT task scheduling, hybrid methods are always employed. The hybrid approaches utilize design-time computations to identify a timing constraint satisfying allocation at runtime [36,37].

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, sec. 4.1, printed p. 6).

Local coordination

Figure 4: Taxonomy of local coordination in MRS (Liu et al. 2018)

• System model (based on the 3-phase look–compute–move cycle):
  • Fully-synchronous: All robots must be in the same phase at the same time.
  • Semi-synchronous: All active robots must be in the same phase.
  • Asynchronous: Robots have no relationship concerning the phases.
• Robot ability:
  • Visibility:
    • Unlimited range: All robots can see each other. The system typically employs a unit disk graph as the visibility graph when limited visibility range is considered.
    • Solid: Robots cannot be looked through.
    • Transparent: Robots can be looked through.
  • Memory model:
    • Oblivious: Robots forget everything once a cycle is finished.
    • Unbounded: Robots can memorize all past information and computations for future use.
    • Finite-state: Robots have a constant, persistent memory.
  • Geometric agreement:
    • Consistent compass: Robots share the same coordinate system.
    • Chirality: Robots agree only on the direction of one axis.
    • Disorientation: Robots have no common sense of the coordinate system.
• Problem:
  • Gathering: Congregating a set of robots, initially placed at arbitrary positions, to a single location within a finite time.
  • Pattern formation: Arranging a set of robots to form a given pattern.
  • Flocking: Moving a group of robots while maintaining a given pattern.

Source section: 4.2. Local Coordination in Multi-Robot Systems

Local Coordination is critical for multi-robot applications. It is also a research hotspot in the field of distributed computing. In literature, the researchers have established various theoretical models for MRS. Based on those models, the fundamental coordination problems are studied. Specifically, the theoretical models are divided into fully-synchronous, semi-synchronous, and asynchronous [38] and fundamental coordination problems include gathering [39], pattern formation [35,40], flocking [41], and so on. We categorize the local coordination in MRS according to three criteria, namely, system model, and robot ability, the problem as shown in Figure 3. Concerning coordination in MRS, a single robot is often modeled as a loop to perform the three stages of the Look–Compute–Move [38]. In the phase of “Look”, the robot acquires data from itself, surrounding robots and the environment. In the “Compute” phase, the robot runs a deterministic program designed to calculate the action to be taken with the data collected in the “Sense phase. Finally, in the”Move” phase, the robot performs the actions generated by the “Look” phase. Based on the model of “Look–Compute–Move”, there are three theoretical models, namely fully-synchronous, semi-synchronous and asynchronous. For a fully-synchronous multi-robot system, all robots must be in the same phase at the same time. In other words, if there is a robot in the “Compute” phase, then it can be asserted that all other robots are in the “Compute” phase. For a semi-synchronous multi-robot system, all robots that are active must be in the same phase. For an asynchronous multi-robot system, the robots do not have any relationship concerning the phases. The research community always wants to achieve local coordination with the weakest robot ability. By saying robot ability, it includes visibility, memory, geometric agreement, and so on. Concerning visibility, the robots can have unlimited or limited visibility range and can be solid or transparent. With unlimited visibility range, all the robots can see each other while the system always employs unit disk graph as the visibility graph when limited visibility range is considered. The robots are solid if the robots cannot be looked through while transparent robots can. Concerning memory, three models are considered, namely oblivious, unbounded memory, and finite-state. In the oblivious model, the robots forget everything once a cycle is finished. In the unbounded memory model, the robots can memorize all the past information and computation for usage while finite-state model means that the robots have a constant persistent memory. As for the ability of geometric agreement, three models consistent compass, chirality, and disorientation are often considered. In the model of consistent compass, all the robots share the same coordinate system; in the model of chirality, the robots agree on only the direction of one axis; and, in the model of disorientation, the robots have no common sense on the coordinate system. The gathering problem refers to congregating a set of robots, which are initially placed at arbitrary positions, to a single location within a finite time. The gathering problem is the first problem studied in multi-robot coordination and is still a research hotspot. In addition, it has been studied in fully synchronous, semi-synchronous and asynchronous models [42,43]. The pattern formation problem refers to arrange a set of robots to form a given pattern. In literature, some research works are focusing on some specific patterns, such as circular [44], regular polygonal [45,46], etc. In addition, some research works are focusing on arbitrary patterns [47,48]. The flocking problem is a more complicated task than pattern formation. It studies how a group of robots move while maintaining a given pattern. In nature, the flocking behavior exhibits in many living beings such as fish and birds. According to whether there are leaders (such as the bellwethers in a group of wild geese), the flocking problem is divided into guided flocking [49] and homogeneous flocking [50].

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, sec. 4.2, printed pp. 6–7).

Human activity recognition

• Human activity recognition + smart warehouse = intelligent services.
• CPS¹ will also focus on monitoring physical and physiological information of humans:
  • remote operations involving humans and machines.
  • worker fatigue, illness, or injury.
  • sorting and delivery of goods to avoid product-classification errors.
  • abnormal behavior by specific or unauthorized individuals.

Table 1: Human activity-recognition approaches (Liu et al. 2018).

	Devices	Advantages	Limitation
Vision-based	Camera (e.g., Kinect, web camera)	Precise capture of human behavior	Vulnerable to bad lighting conditions
Sensor-based	On-body sensors (e.g., accelerometer and gyroscope)	Wide popularity of smart devices	Intrusive to wear sensors
RF-based	RF transceivers (e.g., Wi-Fi router and RFID)	No need to wear sensors (nonintrusive)	Vulnerable to environmental changes

Many areas in the industry can benefit from human activity recognition. For manufacturing, human activity recognition can be applied for the convenience of workers, like remote operations between human and machines. In addition, it can also be used for monitoring the tiredness of workers in the case of getting sick or injured. For transportation, human activities related to the sorting and delivering of goods can be detected and monitored to avoid mistakes on the products classification. In terms of management, abnormal behavior of specific or unlicensed individuals can be recognized if they are going to perform damages to anything in a smart warehouse.

In the past few decades, different kinds of technologies have been brought up for human activity recognition. Most of the existing work on human activity recognition can be classified into the following categories: (1) vision-based human activity recognition; (2) sensor-based human activity recognition; and (3) RF-based human activity recognition. Some of them have already been applied in some industries. Next, we will mainly introduce and give a thorough summary of the above three approaches.

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, sec. 5, pp. 9–10).

For vision-based human activity recognition, there are mainly three steps: (1) Human Segmentation; (2) Feature Extraction; and (3) Activity Recognition. These three procedures are the basis for vision-based human activity recognition. There are higher level scenarios and applications that require specific setting and modeling, for instance, single-person or multi-person scenario, gesture recognition, human identification, etc.

The first step, human segmentation, is to segment the human object from the images or a sequence of video screen-shoots. Common methods for human segmentation is to apply background segmentation [51,52] on the images and foreground human objects are extracted. Then, feature extraction is performed to describe the characteristics and movements of the segmented human objects. Different categories of features can be leveraged as features, including space-time volume [53,54], frequency [55], local descriptors [56] and body modeling [57,58]. The last step relies on the activity recognition and classification algorithms to recognize different kinds of human activities based on the features extracted above. Traditional, Dynamic Time Warping [59], Hidden Markov Model [60] and some machine learning algorithms, e.g., SVM (Support Vector Machine) and ANN (Artificial Neural Network), are applied for activity recognition. In recent years, with the development of deep learning technologies, activity recognition has been realized with greater results. For example, Recurrent Neural Network [61] and Convolutional Neural Network [62] have been applied in pose recognition.

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, sec. 5.1, p. 10).

Sensor-based human activity recognition is to directly attach some sensors on human’s body for activity recognition. Wearable sensors, such as, accelerometer, gyroscope, thermometer and heart monitor, are attached to the body to obtain the information of human motion [63], temperature [64] and health condition [65], etc. Some sensors have been encapsulated into the mobile phone or smartwatches, so people can just use those devices for activity recognition, which can be quite convenient.

The general workflow of sensor-based human activity recognition is as follows: (1) Data Collection; (2) Data Preprocessing; (3) Feature Extraction; and (4) Activity Recognition. Sensors and attributes are selected to collect data during data collection. Since there are some noises or redundancy in the raw data of the collected data, we need to calibrate the sensor data for further use. Concerning the step of feature extraction, many works tend to extract the features defined by themselves in which features can be extracted and selected one by one statistically and analytically, while people are now using the deep learning algorithm to learn the features automatically without giving too much interpretation or explanation of the properties of the extracted features [66]. However, there is a trade-off between the recognition performance and computation resources when applied different learning algorithms on activity recognition. For example, if mobile phones are used for activity recognition by applying deep learning algorithm, since the computation costs a lot, although the result is decent, the battery will soon run out.

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, sec. 5.2, pp. 10–11).

RF-based Human Activity Recognition is to leverage the RF signals, e.g., WiFi, RFID, and Radar signals to recognize a set of activities performed within the detection range of RF signals. The key insight that RF signals can be applied for human activity recognition is that the human presence could affect the propagation of the wireless signal from the transmitter to receiver. In addition, different activities can have different effects on the RF signals so that the pattern in the received signals could be distinctive for different activities. Compared with vision-based and sensor-based approaches, people do not need to wear any sensors on the body, and no pictures will be taken that protect people’s privacy to some degree.

RF-based human activity recognition can be classified into two categories, namely data-based approaches and model-based approaches. Data-based approaches mainly rely on the discovering of the pattern from the data or extracted features of the RF signals based on the machine learning algorithms [67–71]. Some of the extracted features can be interpreted by the behavior while some of them may be hard to explain, e.g., those apply deep learning to extracted features. We do not need to build complex models to describe the behavior; we can learn from the data, so it has broad application scenarios. However, large amounts of training data sets are required to be collected, which is quite labor-intensive. In addition, the wireless signals are sensitive to the environmental changes. So, the training sets will lose its effectiveness when the environment changes.

Model-based approaches are proposed to address the limitations of data-based approaches [72–74]. For human activities, some physical quantities of the behavior will be extracted through the wireless communication models, such as angle of arrival (AoA), time of flight (ToF), and speed. Human activities can be interpreted by the physical quantities and then be recognized. However, many commercial devices do not support extracting the precise physical quantities directly from the devices. Therefore, people either need to pay a lot for expensive and specialized devices or perform calibration on the RF signals. Meanwhile, since some activities could be too complicated to be modeled, model-based approaches have limited application scenarios.

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, sec. 5.3, p. 11).

The general architecture for human activity recognition can be summarized as shown in Figure 4. In the future, not only the object monitoring but also the physical and physiological information of human will be the focus of cyber-physical systems, and, for the industry 4.0, human activity recognition will be integrated to each part of the smart warehouse to provide more intelligent services.

To summarize, the above three approaches for human activity recognition all have their advantages and limitations, as listed in Table 2. Vision-based approaches show encouraging performance for human activity recognition. However, there are still many challenging problems for real-world deployment. Using wearable sensors is a straightforward way for human activity recognition. Ever since the person is performing any activity, there will be a signal reflecting the behavior. However, since people have to wear different kinds of sensors on their body, it could be intrusive for the users. RF-based human activity recognition has attracted much attention since our environment has been covered by wireless signals. Thus, we need to leverage the RF signals around expanding its applications. However, there are still many challenging issues for RF-based human activity recognition. There is still a long way to make it deployed for daily and industrial use.

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, sec. 5.3 and table 2, pp. 11–12).

The source’s Table 2 makes the comparison explicit:

Comparison of human activity-recognition approaches (Liu et al. 2018, table 2, p. 12).

Approach	Devices	Advantages	Limitation
Vision-based	Camera, e.g., Kinect and web camera	Precise capture of human behavior	Vulnerable to bad lighting conditions
Sensor-based	On-body sensors, e.g., accelerometer and gyroscope	Wide popularity of smart devices	Intrusive to wear sensors
RF-based	RF transceivers, e.g., WiFi router and RFID	Do not need to wear sensors (non-intrusive)	Vulnerable to environmental changes

“Precise capture” is the source’s comparative table attribute, not a universal accuracy guarantee.

1. CPS: Cyber–physical systems.

Industry 4.0

Figure 5: Popular techniques for Industry 4.0 and research topics in the context of smart warehousing (Liu et al. 2018).

“As stated above, a CPS-based smart warehouse contains four main components: CPS devices, inventories, robots, and human beings. For CPS devices, there can be thousands of CPS devices working together. Thus, we need to schedule the communication among the CPS devices efficiently. For inventories, we want to timely know their status and location context according to the data reported by the attached devices. For robots, they are required to collaboratively accomplish some tasks that are repetitive and harmful to human beings. For human beings, we need to recognize their gestures or other activities to do warehouse operation. The gestures and other activities may be extracted using the CPS devices deployed in the environment.”

— CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies (Liu et al. 2018, 2).

“3 Information interconnection In the last decades, CPS and IoT, which emerged as the extension of computer networks into every device and object, have become essential supporters for enabling the interconnection of logistics activities. CPS and IoT turn the ideal concept into a reality that goods, logistics facilities, and logistics elements are connected and communicated (Ben-Daya et al., 2017; Lee et al., 2018; Yang et al., 2019). In today’s warehouse practice, RFID, warehouse management system (WMS), augmented reality (AR), reinforced learning, and other emerging technologies enhance the interconnection between different processes and entities (Dou et al., 2015; Bottani and Vignali, 2019; Sartoretti et al., 2019). Interconnection technology provides ways to establish warehouse information collection and exchange, which could be a leading factor for future warehouse development. IoT and CPS technologies, including Radio frequency, pick to light, and pick by voice, are commonly used in the warehouse picking process. The brief introduction on interconnection technology could bring some insights for a better understanding of the smart warehouse. To formulate a way of choosing suited interconnection technology, Hassan et al. (2015) identify the affecting factors of warehouse automatic identification technology through a literature review. The major affecting factors for selecting automatic identification technology include organizational, operational, structural, resources, external, environmental, and technological factors. RFID is a technology for automatic object identification and data collection based on radio waves. Among the interconnection technologies, RFID is considered a choice for positioning, identifying, information interaction, and warehouse management in the smart warehouse. Its strengths include non-contact sensing ability, high flexibility, configurability, high efficiency, and low cost (Tao et al., 2014). Giusti et al. (2019) work on a generalized performance comparison model of different RFID implementations. The benefit related to informatization on warehouse performance has been investigated. A probabilistic risk assessment model is applied to identify and quantify the delay value in transfer time. The application of RFID in smart warehouses mainly focuses on inventory tracking, localization, and warehouse management. Information exchange, which is the main function of the RFID system, significantly impacts the warehouses’ efficient communication. Information accuracy is essential for managing warehouse information exchange efficiency and effectiveness (Zhong et al., 2015). Object localization based on RFID technology serves as a major function of RFID warehouse management. One of the most challenging issues in RFID-based applications is the acquisition of location information of either the tags or readers. Thus, RFID-based localization (especially threedimensional localization) has been an active research topic (Lu et al., 2018). The use of tracking inventory is also investigated by researchers (Shahzad and Liu, 2015; Zhou et al., 2017; Mo and Li, 2019; Gareis et al., 2021). Missing tags searching is another RFID object location and inventory tracking application (Chen et al., 2017). IoT and CPS technologies can also influence the way of warehouse design and operation. RFID tags are attached to products at the item level instead of the pallet level, providing a way to support warehouse storage assignment with higher precision (Choy et al., 2017). RFID can provide infrastructure for collaborative warehouse order fulfillment of decentralized management to improve system reaction capabilities (Reaidy et al., 2015). The real-time information exchange and collection of RFID can provide a decision support tool for order fulfillment (Lam et al., 2015). Furthermore, the energy-saving improvement of RFID could be a new point for the warehouse system. In some research about conflict-free routing in the robotized warehouse, CPS is used for information exchange and pre-detection of potential collisions. Collisions can be detected by comparing the occupancy of the time window, and some corresponding collision avoidance strategies can be developed for conflict-free routing (Lee et al., 2019; Keung et al., 2020). WMS is a computerized database application widely used by the logistics section for warehouse control and optimization. A common WMS includes warehouse processes including receiving, storage, order picking, packing, and shipping. IoT provides ways for information collection and exchange for WMS. The implementation of IoT-based WMS can also facilitate warehouse operation. Qiu et al. (2015) introduce a framework of IoT-enabled Supply Hub in Industrial Park (SHIP). The innovations of IoT-enabled SHIP include different information sharing levels, information structure, and decision-making tools. Lee et al. (2018) propose an IoT-based WMS integrated with the fuzzy logic technique. Based on the case data, the IoT-based WMS shows its competitiveness in improving warehouse productivity, picking accuracy, efficiency, and robustness in order variability.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 38).

Equipment automation

• Automation is the technology used to process tasks with minimal manual labor.
• Benefits include:
  • increased efficiency,
  • agile reaction to customer orders, and
  • reduction in labor costs and errors.

Exploring the relationship of strategic and tactical decisions along with some system characteristics is important when applying automation technology to warehouse design and operation.

“Equipment automation

Warehousing is often seen as a labor-intensive activity. Managing the traditional manual warehouse has become a critical problem for performance improvement with great complexity. This problem is addressed by equipping many warehouses with automation equipment. Automation is the technology used to process tasks with minimal manual labor. It changes warehouse operation in many aspects, including increased efficiency, agile reaction to customer orders, and a significant reduction in labor costs and errors. Automated parts-to-picker order picking systems have become popular for warehouse operations. The increase in the application of automation facilities is significant; consequently, finding optimal strategies for the design and control of automated warehouse systems becomes more challenging with the growth and penetration of modern computer technology.

When applying automation technology to warehouse design and operation, exploring the internal relationship of strategic and tactical decisions along with some system characteristics is important. This section will go through various automated equipment used in smart warehouses and identify some system characteristics that can be used as important factors in strategic and tactical decisions in smart warehouses.

AS/RS is a widely equipped warehousing system in warehouses and distribution centers for automatic material handling with high speed and accuracy. It is the combination of automatic equipment and corresponding control policies. Although AS/RS has been widely adopted for years, research on its operational decision is still of great value, providing basic references for various warehouse technologies in smart warehouses. Today, AS/RS has evolved into an efficient system with many variants, including AVS/RS, SBS/RS, compact AS/RS, and puzzle-based storage and retrieval systems. The system configuration of AS/RS and its variants can be affected by many factors, such as the configuration of racks, aisles, cross-aisles, space, tiers, shape, depots, shuttles, and pick positions. Some tactical policies can also influence system design and performance, such as storage policy, dwell point policy, and input/output point policy. Travel time is a key indicator to evaluate system efficiency, which is measured by the maximum of the isolated horizontal and vertical travel times and is thus applied in warehouse strategic and tactical design.

AGV is a driverless material handling equipment for horizontal material movement. AGV-assisted order picking can adapt to order picking of small-sized items and heavy items. Based on AGV technology, AMR and robotized warehouse systems are developed to assist or replace human labor in product picking and handling. AGV-based warehouse systems and robotized warehouse systems are increasingly applied in warehouses.

RMFS, or KIVA system, is an automated, parts-to-picker picking system in which a fleet of robots can move racks (inventory pods) with products toward picking workstations. It shows greater flexibility and productivity through its flexible layout and hardware installation. The key decisions in the RMFS design include the number of robots, the shape of the forward area, the number of pods per stock-keeping unit (SKU), the ratio of the number of picking stations and replenishment stations, and the replenishment level. The travel time and operation cycle of robots in RMFS are also critical considerations. Other types of robots for parts-to-picker picking modes, such as cable-driven parallel robots, rack-climbing robots, and robots with transport and lifting capabilities, are also powerful equipment in smart warehouses.

Warehouse robots in the picker-to-parts mode can perform collaborative picking with human pickers to minimize the pickers’ travel times and distances. Two types of warehouse robots are commonly used: one type can grasp items from the shelf, and another type can deliver items from the picker to the workstation. Robots can also perform sorting of the picked items.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 43).

Strategic decision

• System characteristics must be considered in warehouse design (they affect travel time).
  • Movement characteristics
    • Acceleration/deceleration
    • Speed of the storage/retrieval (S/R) machine
  • Command cycle
    • Single command cycle (SC)
    • Dual command cycle (DC)
  • Battery management
    • Battery charging
    • Battery swapping
  • Picking-station efficiency
    • Handling speed

Strategic decision

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 43).

With respect to travel time in strategic warehouse design, system characteristics play a crucial role as they directly influence the system’s overall performance. These characteristics can vary across different warehouse systems. Below are the key factors to consider:

• Movement Characteristics:
  • Acceleration and deceleration, as well as the speed of storage/retrieval (S/R) machines, significantly impact travel time, especially in scenarios with frequent starts and stops.
  • The speed profile is an important factor in optimal storage rack design.
  • Acceleration and deceleration of shuttles and lifts influence task scheduling.
• Command Cycle:
  • Defines the operation pattern of warehouse systems. Common types include:
    • Single Command Cycle (SC): Separate storage and retrieval operations.
    • Dual Command Cycle (DC): Combines storage and retrieval operations within a single movement, resulting in better performance in most cases.
  • Other operation modes, such as the Quadruple Command Cycle, have also been analyzed.
• Battery Management:
  • Robots and AGVs are battery-powered, and their battery life depends on capacity and working intensity.
  • Common strategies include:
    • Battery swapping.
    • Plug-in charging.
    • Inductive charging, which offers better throughput time but may have higher costs.
  • Reducing charging duration and increasing charging frequency can enhance productivity and flexibility.
  • Mobile chargers can improve the flexibility of the charging process.
• Efficiency of Picking Stations:
  • Handling speeds at workstations influence warehouse design and operation.
  • Optimized robot assignments and appropriate shelf block sizes can improve efficiency.

Tactical decision

• Storage policy
  • Dedicated: Each SKU is assigned a specific storage location, ensuring predictability and ease of retrieval but potentially leading to underutilized space.
  • Class-based: SKUs are grouped into classes based on characteristics like demand or turnover, and each class is assigned to specific zones, balancing efficiency and flexibility.
  • Random: SKUs are stored in any available location, maximizing space utilization but requiring robust tracking systems for efficient retrieval.

4.2 Tactical decision Storage assignment allocates the storage locations to items. It can be a conductive factor to warehouse strategic and operational decisions. Different storage policies can influence the optimal warehouse shape and storage space (Guo et al., 2016). For example, the operational decision, picking position assignment, and workload balance among pickers can also be influenced by the storage policy (Pan et al., 2015; Ramtin and Pazour, 2015). Class-based storage policy is one of the most commonly used storage policies in practice. It divides stored items into different classes according to product characteristics such as demand, product turnover, and life-cycle (Manzini et al., 2015; Yuan et al., 2019). A small number of storage classes in a class-based storage policy is usually optimal (Yu et al., 2015). Under class-based storage policy, different system configurations and first zone boundary can result in different system performances (Zaerpour et al., 2017b; Zaerpour et al., 2017c). All SKUs of the same type are stored in a unique storage position in a dedicated storage policy. Kress et al. (2017) formalize a storage partition problem that subdivides items into disjoint subsets aiming to minimize the number of groups accessed in retrieving process and show its applicability in storage systems with a dedicated storage policy. A shared storage policy allows multiple products to share one storage rack. Retrieval time and storage utilization can be improved through shared storage (Zaerpour et al., 2015). According to Zou et al. (2018a)’s study, in robotic compact storage and retrieval systems, dedicated storage policy performs better than shared policy in terms of throughput time. In contrast, the shared storage policy is more sustainable from a cost perspective.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 43).

Material flow in a typical automated warehouse

Figure 6: Material flow of an automated retail warehouse with typical storage and handling¹ systems. No manual handling is involved, except for the technicians responsible for keeping the system operational (Azadeh et al. 2019).

In such an automated retail warehouse, selected suppliers unload their own trucks and feed the preannounced single–stock-keeping unit (SKU) pallets to a check-in conveyor (Step 1). The pallets are then stored in an AS/R system (Step 2). When a certain product is requested, the pallet is offloaded and automatically destacked (Step 3). The loose cases are then often put on trays to ease manipulation and are stored in a miniload AS/R or in an AVS/R system (Step 4). When the store order arrives, the cases are retrieved and sequenced (Step 5), and mixed-case palletizers build the pallets or roll cages in a store-specific sequence that allows rapid shelving in the store (Step 6). These roll cages then wait in an order consolidation buffer (OCB), usually an AS/R system (Step 7), until the departure truck arrives, after which they are retrieved and loaded in the sequence determined by the stop sequence in the truck route. Apart from the (many) technicians needed to keep the system alive, no manual handling is involved.

— Robotized and Automated Warehouse Systems: Review and Recent Developments (Azadeh et al. 2019, 918).

1. OCB: Order Consolidation Buffer; AVS/R: Autonomous Vehicle Storage and Retrieval

Process integration

• The operations management of the various processes in warehouses is logically interrelated.
• A well-managed warehouse should consider the combinations of storage, batching, zoning, and routing policies.

“Process integration The operations management of the various processes in warehouses is logically interrelated. In a well-managed warehouse, the consideration of interactions among various processes should be put forward for further improvement of warehouse performance. Thus, integrated planning in warehouse operations management should be implemented. van Gils et al. (2018a) reveal the interactions and relations among storage, batching, zoning, and routing policies of the order picking process. The results indicate that the integrated decision of different processes can attain remarkable benefits in warehouse research and operations. van Gils et al. (2018b) further review the combinations of order picking planning problems in warehouse system design. The two reviews highlight the research value of combinational problems in warehouse operations management. This section will go through the research of smart warehouses at the operational level and determine some topics related to the integrated decision-making of smart warehouse operations management. The first two topics are order processing and location assignment, which try to implement overall planning in smart warehouse operation. The other two topics are resource allocation and differentiated service and blocking rearrangement and conflict-free routing, which deal with the new problems raised in smart warehouse operation.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 45).

Order processing

• Order processing includes:
  • Order sequencing: Proper sequencing can reduce unproductive equipment movement and thus reduce total travel time and makespan.
  • Order batching: Scattered demands from different orders are consolidated and fulfilled as a batch.
  • Split order: Order is divided into two or more parts for picking, and different pickers may pick these parts.

“5.1 Order processing Orders released to the system usually contain the demand information and attributes, such as task release time and due date (Man et al., 2021). Orders may be sequenced, split, or batched to consolidate the scattered demand and reduce the unproductive works of order pickers (Emde et al., 2020; Wang et al., 2020c). Order sequencing mainly determines the sequence of orders to be processed. Proper sequencing can reduce the unproductive movement of the equipment and thus reduce the total travel time and makespan (Gagliardi et al., 2015). Foumani et al. (2018) adapt the traveling salesman problem to a robotic AS/RS with a cross-entropy method to find the optimal sequence of orders and optimal sequence of multi-location items inside each order. Order batching is a way to eliminate the inefficiency of piece picking, in which the scattered demands from different orders are consolidated and fulfilled as a batch. A batch is composed of a subset of orders and the set of items corresponding to these orders. The productivity of order batching on system throughput is significant in automated warehouses (Nicolas et al., 2018). The order batching method applied in research is usually fixed batching, where all orders belonging to a batch are collected before the next batch is started. According to Lenoble et al. (2021), rolling batching, where a new order immediately replaces each completed order belonging to the current batch, and other dynamic batching methods can improve system efficiency in terms of throughput time. As the contrast of order batching, the split order means that an order is divided into two or more parts for picking, and different pickers may pick these parts. The power of order splitting in RMFS and the assignment of pods and orders to picking stations are investigated by Xie et al. (2021). Jiang et al. (2021) investigate the integrated decision of order splitting, batching, and batch assignment by constructing a two-stage optimization model. The first stage of the optimization model works out order splitting and batching strategy, and the second stage assigns the batches formed in the first stage to AGVs for order fulfillment. Most of the warehouse picking literature concerns the optimization of the picking process, while considering the picking and replenishment process remains an open topic. A proper order sequencing can facilitate the replenishment process and prevent stock-outs in A-frame system with fewer workers (Boywitz et al., 2019). In RMFS, the decision problem of order picking and batching along with the replenishments could reduce the total shelf visits (Jiang et al., 2020).”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 45).

Retrieval location assignment and scheduling

• Retrieval tasks create two interrelated decisions:
  • Location assignment determines the locations for the items to be retrieved.
    • Example: In RMFS¹, selecting racks for the retrieval items.
      
  • Retrieval scheduling establishes the schedule of warehouse equipment for the retrieval process.
• The trade-off of the two decisions must be carefully considered to optimize the total travel time.

“5.2 Retrieval location assignment and scheduling Warehouse systems need to perform retrieval tasks so that the items stored in warehouses can be accessed. The retrieval tasks raise two interrelated decisions, namely, location assignment and retrieval scheduling. Location assignment is about how to determine the selection of the locations for the retrieval items. Retrieval scheduling establishes the schedule of warehouse equipment for the retrieval process. The two decisions often interact with each other in real-life examples. The trade-off of the two decisions must be carefully considered to optimize the total travel time (Yang et al., 2015a; Heshmati et al., 2019). Besides, proper storage policies, such as shared storage policy, can benefit location assignment and retrieval scheduling (Yang et al., 2015c). In RMFS, the retrieval location assignment problem can be interpreted as selecting racks for the retrieval items. The objective can be either minimizing travel time or minimizing the number of rack visits. Order sequencing and batching can affect the decision of location assignment. Boysen et al. (2017) carry out a typical integrated order processing problem, where order sequencing, rack visits, and the study of the shared storage policy are optimized. Weidinger et al. (2018) propose the problem of rack parking location assignment during order processing in RMFS. An interval scheduling problem is formalized for the problem, and a metaheuristic dubbed adaptive programming is introduced to solve the problem.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 46).

1. RMFS: Robotic mobile fulfillment system.

Resource allocation and differentiated service

• Some resource allocation strategies are applied to robots to enhance system utilization.
  • Dedicated strategy: Assign each robot as a dedicated resource in the system.
  • Pooled strategy: Group available resources to maximize utilization and counteract variability in resource management.
• E-commerce companies are moving toward customer differentiation and prioritization.
  • The differentiation and prioritization in warehouse operations indicate the different classes assigned to orders.
    • The class is usually based on price or customer priority.
    • Orders with higher classes are attached with tighter time limits than the lower-class orders.
  • Warehouses should prioritize the allocation of resources to ensure the fulfillment of higher-class orders.

“5.3 Resource allocation and differentiated service Robots are critical resources in RMFS. Some resource allocation strategies are imposed on robots to enhance system utilization. Resource pooling strategy refers to group available resources together to maximize utilization of resources and counteract variability in the division of resource management. The dedicated strategy in RMFS refers to that a robot appears as a dedicated resource in the system. In contrast, the pooled strategy breaks the property of dedicated resources to allow a robot to process needed tasks. In Yuan and Gong (2017)’s research, two protocols in sharing robots among order pickers are investigated. The advantage of the pooled strategy is presented by reducing the optimal number of robots needed and the throughput time. Roy et al. (2019) compare the efficiency of dedicated and pooled strategies for zone control in RMFS with single and multiple storage zones. The pooled strategy outperforms the dedicated strategy in terms of the expected throughput time of the picking process. However, the replenishment process under the pooled strategy may take a longer time. E-commerce companies are moving toward customer differentiation and prioritization. The differentiation and prioritization in warehouse operations indicate the different classes assigned to orders. The class is usually based on price or customer priority, and orders with higher classes are attached with tighter time limits than the lower-class orders. Warehouses should prioritize the allocation of resources to ensure the fulfillment of higher-class orders. Under this setting, the common objectives for prioritized order picking are to minimize total travel time or distance. In Gharehgozli and Zaerpour (2020)’s research, different priorities are attached to orders based on the order urgency. The research setting of Gong et al. (2021) states that the expedited shipping requests will be immediately processed when a robot is available. Weights can be used to characterize prioritization. In He et al. (2018)’s research, some AGVs are dedicated to certain classes of orders. The weighted latency for different price-based classes of customer orders is optimized. Wauters et al. (2016)’s research minimizes prioritized waiting time of storage and retrieval scheduling problem. The prioritized waiting time is calculated by the weighted sum of waiting time, in which the weights are determined by customer priority.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 46).

Blocking rearrangement and conflict-free routing

• Warehouse operators usually encounter some unproductive activities, such as
  • rearranging blocking items and
  • route congestion.
• These unproductive activities do not add value and can lead to system inefficiency.
• Special strategies should be implemented to avoid unnecessary work.

“Blocking rearrangement and conflict-free routing

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 47).

Warehouse operators usually encounter some unproductive activities, such as rearranging blocking items and route congestion. These unproductive works are not value-added and can lead to system inefficiency. Some special strategies should be implemented to avoid unnecessary works.

The blocking problem usually occurs in a unit-load warehouse or palletized warehouse. The blocking unit-loads may block the requested unit-loads if the storage position or sequence of the unit-loads is misallocated. The blocking unit-loads need to be rearranged and repositioned to a free storage location to retrieve the requested unit-loads. The selection of the free storage locations for blocking rearrangement usually follows the rule of rearrangement of the nearest location, which will reposition the unit-loads to the nearest free storage location. Other rearrangement rules include rearranging to a random location and rearranging to a deterministic point. The rearrangement to the nearest location rule performs the best in terms of expected rearrangement time. In a flow-rack AS/RS, unit-loads are stored on the flow racks’ storage face and slide down to the retrieval face where the retrieving process is processed. The blocking problem in a flow-rack AS/RS may be more troublesome because only the unit-loads on the retrieval face can be directly retrieved. Proper storage assignment and retrieving scheduling can eliminate the possible rearrangement of blocking unit-loads. Robust storage assignment can protect against the uncertainty of due dates to avoid blockings and excessive retrieval effort. Sequencing of storage and retrieval operations with duration-of-stay storage policy can minimize the total travel time and eliminate restoring of blocking unit-loads.

A route in the AGV-assist picking system implies the path that AGV should follow. When performing storage and retrieval operations, the routes of two or more AGVs may conflict, resulting in blocking or colliding. The blocking and collision problem of AGVs may appear within aisles, along the cross aisles, or at the intersections of aisles and cross-aisles. Route intersection must be concerned to avoid collision and unnecessary waiting. The warehouse collision can be classified into four types: head-on collision, cross collision, node-occupancy collision, and shelf-occupancy collision. The head-on collision happens if two AGVs move in the opposite direction on the same path at the same time or pass the same workstation at the same time. The cross collision happens when two AGVs compete for the workstations at the intersections. The node-occupancy collision happens when an AGV stops at a point on another AGV’s incoming route. The shelf-occupancy collision happens when a shelf (rack) stops at a point on another AGV’s incoming route. Based on these classifications, three solution strategies are carried out: selecting the candidate route, waiting for a short period before starting, and modifying the routes. These classifications and solutions provide a framework for conflict-free routing.

Zone control is a simple way to avoid collisions, where a zone (usually an aisle or a zone that contains several aisles) is occupied by only one AGV. Two aisle access policies can prevent head-on collision and blocking of AGVs. The exclusive policy indicates that an AGV must not access an aisle if another AGV is already inside. The parallel policy indicates that AGVs can enter the same aisle simultaneously if they do not cross. The warehouse layout can be simplified as a grid model or abstracted by square topology method in which the ground of the warehouse is divided into many squares. The squares can represent different properties in warehouses, such as aisle and pick-up position, and the driving routes can be expressed as several consecutive squares. Square topology method provides a simplified method to detect the potential route intersection and collision of robots. Based on this, methods for collisions and deadlock prevention based on chains of elementary reservations can be applied. The warehouse layout can be further abstracted as nodes and arcs in the undirected graph. Dynamic resource reservation can be used for scheduling and collision avoidance of AGV warehouses with graph models. Integrated models of conflict-free routing and job shop scheduling can improve the global performance of the system. Coordination of AGVs and robot fleets is a solution for improving production efficiency and flexibility and eliminating congestion and collision in warehouse systems. The system’s control can be decentralized or centralized. Decentralized approaches directly address the complexity of large-scale multi-robot coordination problems, whereas centralized approaches can find the optimal global solutions. Furthermore, holonic scheduling methods can be applied to coordination scheduling.”

Environmental sustainability

• Sustainable logistics is carried out to
  • seek improvements in environmental performance and
  • gain a competitive edge in an eco-friendly way.
• Smart warehouses usually consume more energy than traditional manual warehouses.
• Sustainable warehouses aim to pursue sustainability goals, such as
  • energy efficiency,
  • greenhouse gas emission reduction,
  • cost reduction, and
  • throughput time minimization.
• Sustainability is a common thread running through strategic, tactical, and operational decisions of the sustainable warehouse.

“6 Environmental sustainability Climate change, resource shortage, and other environmental problems have posed significant threats to the future of human society. The term”sustainability”, which progressively incorporates social, environmental, and economic responsibilities, is primarily about the development process that meets the present needs without compromising the ability of the future. In the logistics section, sustainable logistics and supply chain management is carried out to seek improvements in their environmental performance and gain the competitive edge in an eco-friendly way (Dadhich et al., 2015; Yu et al., 2016; Shi et al., 2019). Smart warehouses usually consume much more energy than traditional manual warehouses because more automated facilities are used in smart warehouses. Increasing energy consumption in warehouses becomes a major component of warehouse operational costs. Meanwhile, the increase in energy consumption produces extra carbon emissions to nature, which is the main cause of climate change. Sustainable warehouses aim to pursue sustainability goals, such as energy efficiency, greenhouse gas emission reduction, cost reduction, and throughput time minimization. Sustainability is a common thread running through strategic, tactical, and operational decisions of the sustainable warehouse. In this section, we will summarize and reveal how the smart warehouse can achieve sustainability goals in warehouse operations management. Similar to the division of previous sections, the literature related to sustainable topics in smart warehouses is reviewed and categorized into two aspects: Strategic and tactical decision and operational decision.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 48).

Sustainable warehouse strategic and tactical decision

• Energy consumption and carbon emissions are major factors in sustainable warehouse design.
• Sustainability factors may be considered in energy-consumption-aware evaluation for automated systems (e.g., AS/RS, AGVs, shuttles).
• Proper storage assignment can improve picking efficiency and reduce energy consumption.

“6.1 Sustainable warehouse strategic and tactical decision

Sustainable warehouse strategic and tactical decisions focus on reducing energy consumption and carbon emissions, which are critical factors in warehouse design. Cold-chain logistics, essential for maintaining the freshness of food products, often rely on refrigerated automated storage and retrieval systems (AS/RS). These systems are more adaptable to cold environments than manual labor. In such systems, refrigeration accounts for a significant portion of energy consumption, making energy efficiency a key design consideration. Sustainable design models for refrigerated AS/RS optimize rack configurations and cold cells to minimize total yearly costs, including energy consumption and greenhouse gas emissions.

Compared to traditional crane-based systems, automated guided vehicles (AGVs) and shuttles offer higher flexibility, efficiency, and better environmental impact. These systems achieve a balance between economic and environmental performance. Energy regeneration, such as capturing energy during vehicle and lift deceleration, is another factor influencing sustainable design. Multi-objective approaches can optimize both cycle time and energy consumption, with system design parameters like the number of aisles, bays, and tiers playing a significant role.

The concept of energy-efficient design extends to other facilities, such as live-cube systems, which offer advantages in investment, operational costs, and energy consumption. For tactical decisions, proper storage assignments can enhance picking efficiency and reduce energy usage. Decentralized storage policies based on turnover rates can further optimize energy consumption in automated warehouses.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 48).

Sustainable warehouse operational decision

• Sustainability goals can be supported through warehouse operational decisions involving
  • sequencing,
  • scheduling, and
  • power load management (i.e., avoiding electricity peaks).

Win–win of environmental and operational objectives can be achieved because eliminating useless work can improve efficiency while reducing energy consumption.

“6.2 Sustainable warehouse operational decision

The sustainable goal can be realized in warehouse operational decisions through various methods, such as sequencing, scheduling, and power load management. To some extent, the win–win of environmental and operational objectives can be achieved in the sustainable warehouse because eliminating useless work can improve efficiency while reducing energy consumption. Sequencing and scheduling optimization can contribute to energy saving and greenhouse gas reduction. Dynamic sequencing methods can improve greenhouse gas efficiency in unit-load multiple-rack AS/RS. A two-level optimization framework for dual shuttle cranes scheduling problems can minimize total cost and energy consumption. The upper level optimizes class-based storage location assignment and inventory considerations, while the lower level focuses on multi-period scheduling of cranes, including reward scheduling and workload balance.

Sequencing and scheduling in the sustainable warehouse can also be extended to power-load management. Electricity suppliers often charge higher prices during peak periods and lower prices during valley periods. Power-load management is a customer-side system used to avoid electricity peaks while making better use of electricity valleys. It sets thresholds to determine whether electric consumption during peak periods needs to be avoided. If thresholds are exceeded, some tasks are delayed, and machines stop to reduce power usage. In multi-aisle mini-load AS/RS operations, electricity peaks can be avoided with minimal loss of throughput. In AGV picking systems, power-load management can be implemented by carefully scheduling the battery charging process to avoid higher electricity prices during peak times. The charging schedule can also consider the availability of cheaper renewable energy sources, determining the start time of charging and the assignment of batteries to chargers.”

— A Literature Review of Smart Warehouse Operations Management (Zhen and Li 2022, 48).

References

Azadeh, Kaveh, René De Koster, and Debjit Roy. 2019. “Robotized and Automated Warehouse Systems: Review and Recent Developments.” Review. Transportation Science 53 (4): 917–45. https://doi.org/10.1287/trsc.2018.0873.

CMS. 2018. UK Real Estate Report – Warehouses Unpacked: Box Clever. CMS. https://cms.law/content/download/360096/file/Real%2520estate%2520industry%2520research%25202018.pdf.

Coyle, John Joseph, C. John Langley, Robert A. Novack, and Brian J. Gibson. 2020. Supply Chain Management: A Logistics Perspective. 11th edition. Cengage.

Cross, Melinda, and Joel Duncan. 2024. “UK Industrial Spotlight - Feb 24.” JLL, February 5. https://www.jll.com/en-uk/guides/uk-industrial-spotlight-feb-24.

Fabric. 2019. “Fabric, Formerly CommonSense Robotics, Launches Triangular Micro-Fulfillment Operation to Fulfill on-Demand Grocery Orders as Flexibility in Automation Becomes Paramount.” December 18. https://www.getfabric.com/newsroom/fabric-formerly-commonsense-robotics-launches-triangular-micro-fulfillment-operation-to-fulfill-on-demand-grocery-orders-as-flexibility-in-automation-becomes-paramount.

Gartner. 2021. “Gartner Says Adoption of Hyperautomation by Gen z Will Drive Supply Chain Autonomy.” March 10. https://www.gartner.com/en/newsroom/press-releases/2021-03-10-gartner-says-adoption-of-hyperautomation-by-gen-z-will-drive-supply-chain-autonomy-.

Liu, Xiulong, Jiannong Cao, Yanni Yang, and Shan Jiang. 2018. “CPS-Based Smart Warehouse for Industry 4.0: A Survey of the Underlying Technologies.” Computers 7 (1, 1): 13. https://doi.org/10.3390/computers7010013.

Richards, Gwynne. 2018. Warehouse Management: A Complete Guide to Improving Efficiency and Minimizing Costs in the Modern Warehouse. 3rd Edition. Kogan Page Ltd.

TIMOCOM GmbH. 2019. “Als er in het magazijn niet alleen goederen te vinden zijn, maar ook winst verstopt zit.” December 12. https://www.timocom.nl/bedrijf/newsroom/servicebericht-lege-plekken-kunnen-winst-opleveren-391016.

Wincanton plc. 2020. Wincanton Plc Annual Report and Accounts 2020. Wincanton plc. https://www.wincanton.co.uk/download_file/force/93/248/.

XPO Logistics, Inc., and Nestle. 2021. “XPO Logistics and Nestle Unveil UK Digital Distribution Warehouse of the Future.” March 30. https://investors.xpo.com/news-releases/news-release-details/xpo-logistics-and-nestle-unveil-uk-digital-distribution/.

Zhen, Lu, and Haolin Li. 2022. “A Literature Review of Smart Warehouse Operations Management.” Frontiers of Engineering Management 9 (1): 31–55. https://doi.org/10.1007/s42524-021-0178-9.