Warehouse robot
Last reviewed
Sources
27 citations
Review status
Source-backed
Revision
v3 ยท 4,635 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
27 citations
Review status
Source-backed
Revision
v3 ยท 4,635 words
Add missing citations, update stale details, or suggest a clearer explanation.
A warehouse robot is an AI-driven machine that automates the movement, storage, picking, sorting, and inventory management of goods inside warehouses, distribution centers, and fulfillment facilities. The category spans automated guided vehicles (AGVs), autonomous mobile robots (AMRs), goods-to-person systems pioneered by Kiva, robotic picking arms that use computer vision to grasp items, and automated storage and retrieval systems, all coordinated by AI software for navigation and fleet orchestration. The scale is led by Amazon: by mid-2025 the company had deployed more than one million robots across over 300 facilities worldwide and launched a generative AI foundation model, DeepFleet, to coordinate them. [1][3][4] The rapid growth of e-commerce and global supply chains has driven this adoption, as operators race to increase throughput, reduce costs, and improve worker safety.
A warehouse robot performs material-handling tasks that were historically done by people walking, lifting, and carrying inside a fulfillment facility. The defining shift over the past decade is the move from fixed, infrastructure-bound automation to AI-driven systems that perceive their surroundings, plan their own paths, and coordinate as fleets. Modern systems combine onboard sensors, computer vision, and machine learning so that a robot can localize itself, avoid obstacles, recognize and grasp a previously unseen item, and re-plan around congestion in real time. Because warehouse associates can spend as much as half of their working time simply walking during order picking, automating that travel is one of the largest available productivity gains in logistics. [22]
The concept of automated material handling in warehouses dates back to the mid-20th century. The first automated guided vehicles (AGVs) appeared in the 1950s, using wire-guided navigation to follow fixed paths along factory and warehouse floors. These early systems were expensive, inflexible, and limited to large manufacturers. Through the 1980s and 1990s, AGV technology improved with the introduction of laser-guided navigation and magnetic tape guidance, but adoption remained modest and concentrated in automotive and heavy manufacturing sectors.
The rise of the internet and online retail in the late 1990s and early 2000s created new pressure on warehouses to fulfill orders faster and more accurately. Traditional pick-pack-ship workflows, in which human workers walked miles through massive facilities each shift, became a bottleneck. This environment set the stage for a new generation of warehouse robots.
In 2003, Mick Mountz, a former employee of the online grocery startup Webvan, co-founded a company called Distrobot (later renamed Kiva Systems) alongside Peter Wurman and Raffaello D'Andrea. Mountz concluded from Webvan's failure that existing material handling systems were too inflexible and too costly for the demands of e-commerce fulfillment. His solution was to invert the traditional warehouse model: instead of workers walking to shelves, the shelves would come to the workers.
Kiva Systems developed squat, orange mobile robots that navigated warehouse floors using barcode stickers affixed to the ground. The robots slid underneath portable shelving units, lifted them, and carried them to human pick stations. This "goods-to-person" approach dramatically reduced walking time and increased order throughput. Kiva sold two robot models: a smaller unit roughly two feet by two-and-a-half feet, capable of lifting 1,000 pounds, and a larger pallet-carrying model rated for 3,000 pounds.
Kiva's technology attracted customers including Staples, Walgreens, and Gap before a much larger buyer entered the picture.
In March 2012, Amazon announced its acquisition of Kiva Systems for $775 million, at the time the company's second-largest acquisition ever. [1] Amazon integrated Kiva robots into its fulfillment centers at a rapid pace. In August 2015, the company rebranded Kiva Systems as Amazon Robotics LLC and stopped selling the robots to outside customers, effectively removing Kiva's technology from the open market. [1][2]
This move had a ripple effect across the logistics industry. Companies that had relied on Kiva Systems for their warehouse automation were left without a supplier, creating what analysts called a "technology gap." The void spurred a wave of new robotics startups, including Locus Robotics, 6 River Systems, and Fetch Robotics, all founded by engineers with experience in warehouse automation or connections to the Kiva ecosystem.
By June 2019, Amazon had deployed more than 200,000 robots in its facilities. [2] That figure reached roughly 750,000 by 2023 and crossed the one million mark in mid-2025, when Amazon delivered its millionth robot to a fulfillment center in Japan. [3][4]
Warehouse robots can be broadly classified into several categories based on their navigation method, form factor, and function.
AGVs are self-operating mobile robots that follow predetermined paths defined by physical infrastructure embedded in or placed on the warehouse floor. Guidance methods include magnetic tape, painted lines, RFID chips, and embedded wires. AGVs perform well in cost-sensitive, high-volume, and predictable environments where routes rarely change. Common AGV tasks include pallet transport, towing trailers of goods between zones, and feeding assembly lines.
Because AGVs rely on fixed navigation infrastructure, reconfiguring their routes requires physical changes to the facility. This limitation makes them less suitable for dynamic environments with frequently changing layouts.
Autonomous mobile robots represent a significant step beyond AGVs. AMRs use onboard sensors, cameras, and SLAM (Simultaneous Localization and Mapping) algorithms to navigate dynamically without relying on fixed infrastructure. They can detect and avoid obstacles, reroute around congestion, and adapt to changes in warehouse layout in real time.
AMRs are widely used for order picking assistance (collaborative picking), goods-to-person delivery, zone-to-zone transfers, and flexible sortation. Leading AMR providers include Locus Robotics, 6 River Systems (now Ocado Intelligent Automation), Geek+, and Quicktron.
Stationary or semi-mobile robotic arms handle tasks that require dexterity and precision, such as picking individual items from bins, packing orders into boxes, palletizing cases, and depalletizing incoming shipments. These arms use advanced end-of-arm tooling (EoAT), including suction grippers, mechanical fingers, and soft grippers, to handle a wide variety of product shapes and sizes.
Computer vision and deep learning have substantially improved the capabilities of warehouse robotic arms in recent years. Systems can now identify and grasp previously unseen items using AI-driven perception, making them practical for the enormous product variety found in e-commerce fulfillment.
Goods-to-person (GTP) systems are a workflow paradigm rather than a single robot type. In a GTP setup, robots (often AMRs or specialized shuttle systems) retrieve inventory containers or shelving units and deliver them to stationary human pick stations. The worker selects the required items and the robot returns the container to storage. This approach eliminates the majority of walking time and significantly boosts picking productivity: manual pickers typically average 60 to 100 picks per hour, whereas GTP workflows can raise that rate to roughly 250 to 400 picks per hour per station. [22]
Kiva Systems pioneered the GTP concept, and it remains the foundation of Amazon's fulfillment model. Other GTP implementations include Exotec's Skypod system, which uses 3D-mobile robots that climb racking structures at speeds up to 4 meters per second, and Geek+'s PopPick stations. [17]
AS/RS solutions use cranes, shuttles, or climbing robots to store and retrieve goods from dense, multi-level racking. These systems maximize vertical storage space and are common in facilities with high inventory density. Symbotic, which partners with Walmart across all 42 of its U.S. regional distribution centers, uses fleets of autonomous robots within a high-density storage structure controlled by AI-powered software. [18]
Sorting robots automate the process of routing packages or items to the correct destination bin, chute, or conveyor. They are especially common in parcel distribution and returns processing. Amazon's Robin and Cardinal systems handle package sorting in fulfillment and sort centers. [2]
Amazon Robotics operates the largest fleet of warehouse robots in the world and has developed several purpose-built systems since acquiring Kiva Systems. [3]
The Hercules robot is the latest evolution of the original Kiva drive unit. It carries portable shelving pods weighing up to 1,250 pounds across fulfillment center floors, delivering inventory to human pickers. Hercules robots make up the overwhelming majority of Amazon's one-million-plus robot fleet. [3]
Unveiled in June 2022, Proteus is Amazon's first fully autonomous mobile robot designed to operate safely alongside human workers without the need for caged-off areas. [5] Earlier Kiva-derived robots operate in restricted zones separated from employees. Proteus uses advanced perception, navigation, and safety systems to move through shared spaces. It emits a green light beam ahead of it while in motion and stops automatically if a person steps into its path. Proteus transports wheeled carts called GoCarts in outbound handling areas of fulfillment and sort centers. [5]
Introduced in November 2022, Sparrow is Amazon's first robotic system capable of detecting, selecting, and handling individual products from inventory. [6] Leveraging computer vision and artificial intelligence, Sparrow uses a suction-based gripper with multiple vacuum tubes that can be selectively engaged depending on the size and shape of the item being picked. At launch, Sparrow could handle roughly 65% of Amazon's sortable inventory of more than 100 million products. [6] By 2025, Amazon reported that its latest Sparrow handles over 200 million unique products of varying shapes, sizes, and weights. [23]
Sequoia is a multi-level inventory management system that combines robotics, AI, and computer vision to consolidate and store inventory. It enables Amazon to identify and store inventory up to 75% faster than previous methods. [23] The Sequoia system was first deployed in Houston, Texas, and later scaled to Amazon's most automated facility in Shreveport, Louisiana, where it holds more than 30 million items, five times the capacity of its initial deployment. [8][23]
Announced in May 2025, Vulcan is Amazon's first robot equipped with a sense of touch. [7] It features two robotic arms: one that rearranges items within storage compartments and another that uses a camera and suction cups to pick and place goods. Force feedback sensors allow Vulcan to detect the weight, shape, and orientation of packages. Amazon targets a stowing rate of 300 items per hour at 80% item coverage, with the robot operating up to 20 hours per day. [7] Vulcan entered operation at sites in Spokane, Washington and Hamburg, Germany, with broader rollout planned across Europe and the United States. [7]
Amazon's fulfillment center in Shreveport, Louisiana represents the company's most advanced integration of robotics. The facility deploys ten times more robotics than a typical Amazon fulfillment center and uses AI to direct eight different robot models, including Proteus, Sparrow, and Sequoia working in concert. [8][23] Amazon reports that the Shreveport site can process customer orders 25% faster and at 25% lower cost compared to other buildings. [8]
The warehouse robotics industry includes a mix of established automation firms, venture-backed startups, and technology conglomerates. The following table summarizes major players in the field.
| Company | Headquarters | Founded | Key products / systems | Notable details |
|---|---|---|---|---|
| Amazon Robotics | Westborough, MA, USA | 2003 (as Kiva Systems) | Hercules, Proteus, Sparrow, Sequoia, Vulcan, DeepFleet | Over 1 million robots deployed across 300+ facilities [3][4] |
| Locus Robotics | Wilmington, MA, USA | 2014 | LocusBot AMRs | 350+ deployment sites; surpassed 6 billion units picked in 2025 [24] |
| 6 River Systems (Ocado) | Waltham, MA, USA | 2015 | Chuck collaborative AMR | Founded by ex-Kiva executives; acquired by Shopify (2019, $450M), then by Ocado (2023, $12.7M) [13] |
| Berkshire Grey | Bedford, MA, USA | 2013 | AI-enabled picking, sorting, packing systems | Acquired by SoftBank in 2023 for approximately $375 million [14] |
| Fetch Robotics (Zebra) | San Jose, CA, USA | 2014 | Cloud-managed AMR platforms | Acquired by Zebra Technologies in 2021 for $290 million; Zebra winding down AMR division as of 2025 [12] |
| Geek+ | Beijing, China | 2015 | Goods-to-person AMRs, sorting robots | 30,000+ robots deployed; 500+ customers in 40+ countries |
| Quicktron | Shanghai, China | 2014 | C56, M5 goods-to-person robots | 35,000+ robots deployed; strategic investment from KION Group [20] |
| GreyOrange | Roswell, GA, USA | 2012 | Ranger robots, GreyMatter software | 6,000+ robots deployed; $472 million in total funding |
| Symbotic | Wilmington, MA, USA | 2007 | AI-powered AS/RS | Deploying across all 42 Walmart U.S. distribution centers; acquired Walmart's Advanced Systems and Robotics business in 2025 [18] |
| Exotec | Croix, France | 2015 | Skypod 3D-mobile robot AS/RS | 10,000+ robots at 200+ sites; surpassed $1 billion in sales in 2024 [17] |
AI is the enabling technology behind the most significant advances in warehouse robotics over the past decade. The main roles are perception and grasping, SLAM-based navigation, fleet orchestration, and demand forecasting. Key application areas include the following.
Modern AMRs use SLAM algorithms to build and continuously update maps of their environment while simultaneously tracking their own position within that map. SLAM implementations in warehouses typically combine data from LiDAR sensors, depth cameras, and inertial measurement units (IMUs). LiDAR-based SLAM generates point clouds of the surrounding environment, while visual SLAM (vSLAM) uses camera feeds to identify and track visual features. The fusion of multiple sensor modalities provides robust localization even in large, visually repetitive warehouse environments.
Recent advances include semantic SLAM, which uses object detection models such as YOLOv8 to recognize specific objects like shelves and pallets. This allows robots to reason about their environment at a higher level, upgrading path planning from purely geometric navigation to semantically informed decision-making.
When hundreds or thousands of robots share a warehouse floor, coordinating their movements to avoid collisions and minimize travel time becomes a complex optimization problem. Multi-agent path finding (MAPF) algorithms, often enhanced by reinforcement learning, compute collision-free trajectories for entire robot fleets in real time. These systems balance competing objectives such as minimizing total travel distance, reducing congestion at bottleneck areas, and prioritizing urgent orders.
Amazon's fleet management systems coordinate over a thousand robots simultaneously in a single fulfillment center. [3] The algorithms continuously re-plan paths as conditions change, accounting for new orders, blocked aisles, and robots entering or leaving charging stations.
Robotic picking of individual items from unstructured bins remains one of the hardest problems in warehouse automation. Items vary enormously in size, shape, weight, fragility, and packaging. AI-driven pick-and-place systems use convolutional neural networks and other deep learning architectures to segment objects in cluttered scenes, estimate grasp points, and plan arm trajectories that avoid collisions.
Amazon's Sparrow system uses computer vision to identify items among hundreds of millions of unique products and selects the appropriate suction grip pattern. [6][23] Berkshire Grey's systems employ similar AI-driven perception for automated sorting and packing. Force and tactile sensors, as demonstrated by Amazon's Vulcan robot, add another dimension of feedback that allows robots to handle delicate items without damage. [7]
Beyond picking, computer vision plays a role in inventory management, quality inspection, and damage detection. Vision systems mounted on robots or fixed infrastructure scan barcodes, read labels, count items, and flag anomalies. AI models trained on large image datasets can detect damaged packaging, verify product identity, and ensure correct placement within storage systems.
In July 2025, Amazon announced DeepFleet, a generative AI foundation model built specifically to coordinate its robotic fleet. [3][4] DeepFleet acts as an intelligent traffic-management system for robots such as Hercules, Pegasus, and Proteus, and Amazon reports that it improves robot fleet travel time by about 10%, enabling faster deliveries and lower energy costs. [3] The model was trained on Amazon's inventory-movement data and built using AWS tools including Amazon SageMaker. [3] Scott Dresser, Vice President of Amazon Robotics, framed the approach this way: "DeepFleet represents our practical approach to AI innovation. Rather than pursuing technology for its own sake, we're focused on solving real problems." [3]
The integration of generative AI and foundation models into warehouse robotics represents a shift from task-specific models toward more general-purpose robotic intelligence that can understand and adapt to novel situations, such as handling unfamiliar items or adjusting to changes in facility layout.
A new frontier in warehouse automation is the deployment of humanoid robots, bipedal machines with human-like form factors that can navigate spaces designed for people and manipulate objects using arms and hands. As of 2025 and 2026 these deployments are early but growing, led by Agility Robotics, Figure AI, and Tesla.
Agility Robotics developed Digit, a bipedal humanoid robot designed specifically for logistics work. Digit stands roughly five feet tall, has two arms with force-sensing capabilities, and can walk through standard doorways, climb stairs, and operate in environments built for humans. Amazon began testing Digit at its research facility near Seattle in October 2023, initially using the robot for tote recycling (picking up and moving empty totes after inventory has been picked from them). [10]
Digit has since advanced to commercial deployment. Under a multi-year agreement signed with the logistics provider GXO in June 2024, Digit went to work at a GXO facility in Flowery Branch, Georgia (operating for the apparel brand Spanx), where Agility announced in November 2025 that the robot had moved more than 100,000 totes, the first revenue-generating deployment of a humanoid robot in a warehouse. [9] Digit integrates with Agility's Arc fleet-management platform, and Agility Robotics has reported a 98% task success rate after extended testing. [9]
Figure AI developed the Figure 02 humanoid robot, which completed an 11-month deployment at BMW Group's Plant Spartanburg in South Carolina. [11] While this deployment focused on automotive manufacturing rather than warehousing, it demonstrated capabilities directly applicable to logistics: picking sheet-metal parts from racks and placing them on fixtures within a 37-second cycle time. The robots ran 10-hour shifts on weekdays, accumulated 1,250 hours of runtime, and contributed to the production of over 30,000 BMW X3 vehicles. [11] Figure 02 features onboard CPU/GPU processing, a 2.25 kWh battery, a six-camera vision system, and fourth-generation hands with 16 degrees of freedom. [11]
The successor Figure 03 has moved into logistics environments directly. In May 2026, Figure AI ran a continuous autonomous package-sorting demonstration using three Figure 03 robots powered by its Helix-02 vision-language-action model; originally planned for eight hours, the test ran for 81 hours without a mechanical failure, during which the robots sorted 101,391 packages with no human teleoperation. [21][25] Figure AI also signed a commercial agreement with Catalyst Brands to deploy Figure 03 robots at a distribution center in Reno, Nevada, marking the company's first contracted warehouse logistics deployment. [21] Figure's BotQ manufacturing facility reached a production rate of one Figure 03 per hour by mid-2026, a roughly 24-fold increase achieved in under four months, and had delivered more than 350 third-generation units. [26]
Tesla's Optimus (also called Tesla Bot) is a general-purpose humanoid robot that Tesla intends to deploy first in its own factories and eventually in warehouses and other commercial settings. As of early 2026, Optimus Gen 3 units are physically deployed inside Tesla's Fremont and Texas factories, though CEO Elon Musk acknowledged on the Q4 2025 earnings call (January 28, 2026) that the robots are not yet performing useful work and remain primarily used for learning and data collection. [27] Tesla announced plans to repurpose Model S and Model X lines at Fremont to build Optimus, targeting a one-million-unit annual production line by the end of 2026, with a projected production-unit price in the range of $20,000 to $30,000. [27]
Industry analysts expect hundreds to low thousands of humanoid robots to be deployed in industrial settings through 2025 and 2026, with broader adoption still several years away. The humanoid robot market is projected to reach $15.3 billion by 2030, growing at a compound annual growth rate (CAGR) of about 39%. Key challenges include reliability (mechanical failure rates remain higher than traditional robots), battery life, manipulation dexterity, and cost competitiveness against purpose-built alternatives.
The warehouse robotics market has experienced strong growth driven by e-commerce expansion, labor shortages, rising wages, and advances in AI and sensor technology. Market size estimates vary by research firm due to different definitional scopes, but all major analysts project double-digit compound annual growth.
| Source | Market size (2025 or 2026) | Projected size | CAGR |
|---|---|---|---|
| Mordor Intelligence | $10.96 billion (2026) | $24.55 billion by 2031 | 17.5% [16] |
| Fortune Business Insights | $6.51 billion (2025) | $7.35 billion (2026) | 16.8% [15] |
| SNS Insider | $9.79 billion (2025) | $18.73 billion by 2033 | 14.2% |
| Roots Analysis | $8.75 billion (2026) | $32.48 billion by 2035 | 15.7% |
Key growth drivers include the continued expansion of e-commerce (particularly same-day and next-day delivery expectations), persistent labor shortages in warehouse and logistics roles, declining hardware costs for sensors and compute, and the maturation of AI-powered perception and planning software.
As warehouse robots increasingly share space with human workers, safety standards have evolved to address the risks of human-robot collaboration. Key standards include:
Collaborative safety techniques defined in these standards include safety-rated monitored stop, hand guiding, speed and separation monitoring, and power and force limiting. Modern warehouse robots implement these through a combination of LiDAR, 3D cameras, bumper sensors, and real-time fleet management software that monitors the position and velocity of every robot.
The deployment of warehouse robots has sparked debate about their impact on employment. Proponents argue that robots handle the most physically demanding, repetitive, and injury-prone tasks, allowing human workers to focus on higher-value activities such as problem-solving, quality control, and equipment maintenance. Amazon has stated that its robotics investments have created hundreds of thousands of new jobs, including skilled technical roles for robot maintenance and supervision, and that it has retrained more than 700,000 employees since 2019. [3]
Critics point out that automation reduces the number of workers needed per unit of output and that many new roles pay similarly to the positions they replace. Studies from organizations such as the MIT Task Force on the Work of the Future have found that warehouse automation tends to change the nature of work rather than eliminate it entirely, at least in the near term. Workers increasingly collaborate with robots rather than being fully replaced by them.
The collaborative model, in which AMRs assist human pickers by carrying items and optimizing pick routes while humans handle the physical picking, has become the dominant deployment pattern for companies like Locus Robotics and 6 River Systems. This approach can double or triple worker productivity without full displacement. [22]
Despite rapid progress, warehouse robots face several technical and practical challenges:
The warehouse robotics field continues to evolve along several trajectories. AI foundation models trained on diverse manipulation and navigation data, exemplified by Amazon's DeepFleet, promise to make robots more adaptable to new tasks and environments without extensive reprogramming. [3] The convergence of humanoid robotics with warehouse automation could eventually yield general-purpose robots that handle a wider range of tasks than today's specialized systems. Advances in battery technology, edge computing, and 5G connectivity will support longer operating times and faster decision-making. The integration of digital twins, virtual replicas of physical warehouses used for simulation and optimization, is enabling companies to test new robot configurations and workflows before committing to physical deployment.
As consumer expectations for faster delivery continue to rise, the economic incentive to automate warehouses will only intensify, ensuring that warehouse robots remain one of the most active areas of applied robotics and artificial intelligence.