Agility Robotics Digit
Last reviewed
Sources
35 citations
Review status
Source-backed
Revision
v3 · 4,550 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
35 citations
Review status
Source-backed
Revision
v3 · 4,550 words
Add missing citations, update stale details, or suggest a clearer explanation.
Digit is a bipedal humanoid robot built by Agility Robotics, an Oregon-based company that spun out of Oregon State University in 2015. The robot is roughly the size of a small adult, walks on two legs with a distinctive bird-like gait, and is the first humanoid platform to be deployed under a paid commercial contract inside a working warehouse. Digit has been piloted by Amazon since 2023 and is the centerpiece of a multi-year Robots-as-a-Service (RaaS) agreement with logistics provider GXO, where the fleet crossed the 100,000-tote milestone in late 2025. Agility manufactures Digit at RoboFab, a 70,000 square foot factory in Salem, Oregon that the company describes as the first purpose-built facility for mass production of humanoid robots.
Digit sits in an unusual place in the current humanoid race. It is not the most photogenic robot in the field, it does not run a flashy end-to-end neural policy, and its hands are deliberately simple rather than human-like. What it has instead is paying customers and a measurable track record in live operations, which is something most of its competitors, including the Figure 03, the Apptronik Apollo, the Tesla Optimus Gen 3, and the 1X Neo, are still working toward at comparable scale.
The story starts at Oregon State, where mechanical engineering professor Jonathan Hurst ran the Dynamic Robotics Laboratory. Hurst had done his PhD at Carnegie Mellon, where he met Damion Shelton, another robotics graduate student. Hurst's academic work focused on the physics of legged locomotion, in particular how spring-like leg behavior in birds and humans can be reproduced mechanically. That research line produced ATRIAS, a bipedal robot funded in part by DARPA, which served as a testbed for spring-mass walking models.
In 2015 Hurst, Shelton, and OSU graduate Mikhail Jones founded Agility Robotics to commercialize the lab's ideas. Shelton became CEO, Hurst took the chief technology officer role, and the company licensed underlying technology from Oregon State. Early operations were split between Albany, Oregon (next to the university) and Pittsburgh, Pennsylvania (where Shelton was based).
The company's first product was Cassie, a lower-body-only bipedal platform built under a $1 million ARPA-style research contract. Cassie was about the size of a person from the waist down. It had no torso, no arms, and no head, just two ostrich-style legs and a small electronics bay between them. Cassie was sold to research universities starting in 2017 and became a popular legged-locomotion testbed. A team at the University of California, Berkeley taught one Cassie unit to run, and an Oregon State team set a Guinness world record for fastest 100-meter sprint by a bipedal robot in 2022, with Cassie covering the distance in 24.73 seconds.
Digit was unveiled in February 2019 as Agility's second product and the company's first true humanoid. The first robot was about 1.55 meters tall, weighed around 42 kilograms, and added a torso, two arms, and a small sensor head to Cassie's leg architecture. The first two Digit units off the production line were sold to Ford Motor Company, which used them in a research project on last-mile package delivery. Other early customers were universities and research labs.
The initial Digit was clearly a research platform. It had pincer-style end-effectors instead of hands, a limited payload, and a small navigation footprint. Over the next four years Agility produced several major hardware revisions, refining the perception stack, the arm geometry, the battery system, and the controller. By the time the company began shipping commercial units in 2023, Digit had moved from a curiosity in a university hallway to a robot that could pick up a 35-pound warehouse tote and walk across an active fulfillment floor.
Shelton ran Agility for nearly a decade. In March 2024 he stepped down as CEO to focus on public policy and industry engagement, and Peggy Johnson took over the chief executive role. Johnson had previously been CEO of spatial computing company Magic Leap and executive vice president of business development at Microsoft, where she launched the company's M12 corporate venture fund. Johnson's brief was to push Agility from a small pilot-scale operation into a real commercial business, with a focus on sales, partnerships, and ramping production at the Salem factory.
Hurst remains chief robot officer, and the original founding trio is still involved in the company.
Digit is bipedal, with two legs, two arms, a small head with a stereo camera array, and a stubby torso that houses most of the compute, batteries, and connectivity hardware. The legs use a four-bar linkage with reverse-knee geometry, similar to the legs of a bird; the arms have four degrees of freedom each in the current production unit; and the hands are simple two-finger grippers rather than five-fingered dexterous hands.
That last choice is deliberate. Most of Agility's deployed work involves moving warehouse totes, which have standard handles in known locations, so dexterous five-fingered hands would add cost, complexity, and points of failure without unlocking new tasks. The company has talked about more capable manipulators for future versions, but the current commercial unit is optimized around bulk material handling rather than fine-motor work.
Agility has not always given each Digit revision a clear version number in public materials, and the company sometimes describes hardware changes as iterations rather than full generations. The table below collects publicly disclosed specifications for the major releases the company has discussed. Where a figure has not been officially published, the cell is left blank.
| Generation | Year | Height | Mass | Payload | Runtime | Notable changes |
|---|---|---|---|---|---|---|
| Digit v1 | 2019 | ~1.55 m | ~42 kg | ~18 kg | ~1 hour | First humanoid release; lidar head; pincer hands |
| Digit v2 | 2020 | ~1.55 m | ~45 kg | ~18 kg | ~1 hour | Redesigned arms; improved perception sensors |
| Digit v3 | 2021 | ~1.65 m | ~45 kg | ~16 kg | ~1 hour | Warehouse-focused arms; new gripper |
| Digit v4 (commercial) | 2023 | ~1.75 m | ~65 kg | ~16 kg | ~2 hours | First mass-produced unit; humanoid head; tote handling |
| Digit (2025 refresh) | 2025 | ~1.75 m | ~65 kg | ~16 kg | ~4 hours per charge, autonomous docking | New limbs, expanded grasping angles, CAT1 safety stop, FSoE safety bus |
The robot walks at about 1.5 meters per second in normal operation, which is roughly a brisk human walking pace. Top speed on flat ground is slightly higher in benchmark conditions but is rarely used in real deployments. Digit can climb shallow ramps and step over small obstacles, but it does not handle stairs in commercial settings.
Digit's perception stack uses a combination of cameras, depth sensors, and inertial measurement units. The head houses a stereo camera pair and a small lidar module for forward-looking obstacle detection, while the torso and limbs carry additional cameras and force sensors. The robot's onboard compute runs locally for time-critical loops like balance control and immediate obstacle avoidance, while higher-level planning, fleet coordination, and analytics happen in the cloud through the Agility Arc platform.
Motors are custom direct-drive units developed by Agility, with carefully tuned series elastic compliance in some joints. The company has cited bird locomotion as the inspiration for the leg compliance, since the spring-mass model of running is well understood from biomechanics and translates cleanly into low-impedance, energy-efficient walking.
The 2023 commercial Digit shipped with roughly two hours of runtime on a single battery. The 2025 refresh extended runtime to about four hours of continuous operation under the company's published figures, and Agility added an autonomous charging dock so a Digit unit can walk itself to a charger, dock, and return to work without human intervention. Schedules in real deployments tend to overlap charging windows so the fleet covers a full shift even though individual robots are not running continuously.
Digit is rated for indoor environments and is not waterproof. The robot is designed for warehouse, factory, and logistics floors with controlled climate and reasonably flat surfaces.
The 2025 refresh added a Category 1 stop function, safety-rated programmable logic controllers, an on-robot emergency stop button, and Safety over EtherCAT (FSoE) for the internal safety bus. The point of these additions is to bring Digit closer to compliance with functional-safety standards that would let it operate near unprotected human workers rather than only inside a fenced cell. As of 2026 Digit still works in partitioned zones in most deployments, but the safety hardware is the prerequisite for the closer collaboration model the company has been pitching.
On June 22, 2026, NVIDIA announced NVIDIA Halos for Robotics, which it described as the industry's first full-stack safety system for robotics and physical AI, unifying AI compute and functional safety in a single stack. The framework extends the safety approach NVIDIA developed for autonomous vehicles to industrial robots operating in unstructured environments shared with people. Agility Robotics was named as the first company to integrate Halos elements into a humanoid, making Digit the first humanoid robot to adopt the system.[33]
Agility incorporated NVIDIA IGX Thor industrial compute and the Halos software stack into Digit's existing safe human detection system. NVIDIA positions IGX Thor as an industrial-grade platform with embedded safety, pairing high-performance inference with a hardware Functional Safety Island that is rated for IEC 61508 SIL 3 capability and carries on the order of 22,000 safety mechanisms for diagnostic coverage, plus the Holoscan Sensor Bridge for sensor ingest. The Halos architecture is organized in three layers: a platform layer (IGX Thor hardware), a software layer (the Halos Core operating system), and an application layer of safety blueprints and AI agents tuned to specific tasks.[33][34]
A central element of the announcement is the Outside-In Safety Blueprint, a reference architecture that extends a robot's perception beyond its own onboard sensors by using external facility cameras and AI agents to monitor the surrounding space. The blueprint runs a perception pipeline built on NVIDIA Metropolis to detect and track people and equipment across multiple cameras, a safety AI monitor that flags out-of-distribution conditions such as lighting changes or a blocked camera and falls back to onboard safety, an event integrator that fuses multi-camera events with confidence and staleness checks, and a safety decision maker running on the Functional Safety Island that adjusts robot speed or mutes safety constraints based on the integrated picture. In NVIDIA's demonstrated warehouse scenario the system relaxes a safety constraint when no workers are present in a loading area and restores full safety the instant a person is detected.[34]
Alongside the technical integration, Agility joined the NVIDIA Halos AI Systems Inspection Lab, an ANSI National Accreditation Board (ANAB) accredited inspection body that evaluates functional and AI safety. The lab is being used to inspect Digit's safety-related software, AI components, and cybersecurity protections against established and emerging functional-safety standards, including IEC 61508, ISO 13849, and ISO/IEC TR 5469, the technical report that addresses the use of AI in safety-related functions. The intent is to accelerate certification toward the closer human collaboration that Agility has been pitching for Digit. Agility chief executive Peggy Johnson framed the rationale by saying that for humanoids to deliver value at scale, safety has to be built into the robot and validated across the entire system.[33]
NVIDIA's announcement situated the integration in the context of Digit's existing commercial deployments, citing the robot's use in warehouse, manufacturing, and logistics operations with Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada.[33]
Digit's software has two main layers. The first is the on-robot controller, which handles balance, walking, force control, and immediate manipulation. This layer is conventional model-based robotics: the gait controller is grounded in spring-mass locomotion models that Hurst's group worked on at Oregon State, and the manipulation stack is built around classical perception with task-specific learned components for grasp planning.
The second layer is Agility Arc, a cloud platform that orchestrates fleets of Digit robots, monitors performance, integrates with warehouse management systems, and pushes over-the-air software updates. Arc was announced in March 2024. It exposes APIs for integration with warehouse management systems (WMS), warehouse execution systems (WES), and manufacturing execution systems (MES), and surfaces operational KPIs like uptime, throughput, mean time between incidents, and per-task cycle time. The 2025 refresh added remote monitoring and maintenance features so support engineers can diagnose and recover individual robots without site visits.
Agility's software approach is more conservative than the end-to-end vision-language-action policies that companies like Figure and 1X have been promoting. The company has talked about adding more learning-based components, particularly for manipulation and for adapting to new task variants, but the production software has stayed close to a hybrid model-based and learned controller architecture that prioritizes predictability over fluidity. In a humanoid market full of demo videos, Agility has been notably reluctant to publish flashy autonomy demos, and the company is open about the trade-off between making a robot look impressive and making it run safely on a real customer floor.
In 2025 Agility announced a partnership with Manhattan Associates to integrate Digit fleets directly with that company's warehouse management software. The integration lets a Manhattan WMS deployment treat Digit fleets as a managed asset, similar to how it would manage other materials-handling equipment, which lowers the integration cost for customers that already run on Manhattan systems.
Amazon was the first major customer to publicly disclose pilot use of Digit. The retailer announced testing at its BFI1 facility in Sumner, Washington, in October 2023. The site is both a working fulfillment center and an Amazon Robotics research and development location, which gave the team a way to evaluate Digit in real warehouse conditions without disrupting production lines.
Amazon's first task for Digit was tote recycling: picking up empty totes from a conveyor, walking them to a stacking location, and placing them in the correct orientation. The job is highly repetitive, ergonomically rough for humans, and the kind of task humanoid robots can plausibly handle today without needing the kind of fine manipulation that more complex picking would require. Amazon Robotics chief technologist Tye Brady characterized the deployment as early-stage research, and the retailer did not commit to broader rollout at the time.
Amazon's investment in Agility was disclosed as part of Agility's $150 million Series B in 2022, which was led by DCVC and Playground Global with participation from MFV Partners, ITIC, Robotics Hub, Safar Partners, Sony Innovation Fund, TDK Ventures, and Amazon's Industrial Innovation Fund. The investment relationship and the operational pilot are separate but obviously linked.
GXO is a pure-play contract logistics provider that operates warehouses on behalf of brands like Spanx, Nike, and Apple. In June 2024 GXO and Agility signed a multi-year RaaS agreement to deploy Digit at a Spanx fulfillment center in Flowery Branch, Georgia, north of Atlanta. The deal was unusual on two counts. First, it was the first formal commercial contract between a humanoid robot maker and a logistics customer, rather than a pilot or proof-of-concept. Second, it was structured as Robots-as-a-Service, meaning GXO pays an ongoing fee for robot usage rather than buying the units outright.
The RaaS framing is important. It bundles hardware, software, support, integration, and ongoing updates into a single subscription, which lowers the upfront capital cost for the customer and gives Agility a recurring revenue model. GXO chief executive Malcolm Wilson described the agreement as the start of a longer working relationship rather than a one-off purchase, and Agility's then-CEO Peggy Johnson framed it as proof that humanoids can do real warehouse work for a real customer.
Digit's role at Flowery Branch involves moving plastic totes between conveyors and autonomous mobile robots, loading items onto conveyors, and stacking containers at designated floor locations. The work is coordinated through Agility Arc, which talks to GXO's warehouse systems and to the mobile robots already on site.
In November 2025 Agility announced that the Digit fleet at GXO had moved more than 100,000 totes in live commercial operations. The milestone was significant for the humanoid sector because it represented a real volume of paid work in a real warehouse, not a controlled lab benchmark or a curated demo. Coverage in The Robot Report, Robotics and Automation News, and trade press positioned the number as the first concrete proof that a humanoid robot fleet could sustain useful throughput in a paying contract over an extended period.
The figure is also useful context for the rest of the industry. By the same window, no other humanoid maker had published comparable cumulative throughput numbers from a paying customer site. Apptronik's Apollo had moved smaller volumes in Mercedes-Benz pilots, Figure had begun fleet operations at BMW's Spartanburg plant, and Tesla's Optimus had been used in internal logistics tasks at the company's own facilities, but the GXO number set a public reference point for humanoid productivity that competitors are now measured against.
Agility has disclosed additional commercial and pilot relationships:
| Customer | Relationship | Year |
|---|---|---|
| Ford Motor Company | Research purchase of two early Digit units | 2020 |
| Amazon | Pilot deployment at BFI1, Sumner, Washington | 2023 |
| GXO Logistics | Multi-year RaaS contract at Flowery Branch, Georgia | 2024 |
| Schaeffler Group | Minority investment plus humanoid purchase agreement across plant network | 2024 |
| Manhattan Associates | Software integration partnership for warehouse management | 2025 |
| Toyota Motor Manufacturing Canada | Manufacturing and logistics deployment cited in NVIDIA Halos announcement | 2026 |
Schaeffler is a German motion technology company that runs about 100 manufacturing plants worldwide. The November 2024 agreement covered both a minority equity investment in Agility and a multi-unit purchase order for Digit robots, with Schaeffler describing the deployment as part of a plan to integrate humanoid labor across its plant network through 2030. Agility's other customer pipeline, which the company has hinted at in interviews and in materials filed with investors, reportedly includes additional automotive and consumer-goods customers, though specific deployments have not been disclosed publicly. NVIDIA's June 2026 Halos announcement additionally cited Toyota Motor Manufacturing Canada among the operations using Digit.[33]
Agility has raised a series of venture rounds in line with the company's transition from a research-grade legged robotics shop to a commercial humanoid manufacturer. The table below lists publicly reported funding events.
| Round | Year | Amount | Lead investor(s) | Notes |
|---|---|---|---|---|
| Seed | 2015 | undisclosed | ONAMI and early angels | Spin-out from Oregon State |
| Series A | 2018 | $8 million | DCVC, ONAMI | Funded Digit v1 development |
| Series B | 2022 | $150 million | DCVC, Playground Global | Participation from Amazon, Sony, TDK, MFV Partners, others |
| Series B extension | 2024 | undisclosed | Schaeffler | Strategic minority investment alongside purchase agreement |
| Bridge / Series C-2 | 2024 | ~$11.1 million | existing investors | Reported as a smaller follow-on |
| Series C | 2025 | ~$400 million | WP Global Partners, SoftBank | Post-money valuation reported around $2.12 billion |
Cumulative funding through the 2025 Series C is roughly $640 million by public reporting, which puts Agility in the middle of the humanoid funding pack: well above earlier-stage entrants like 1X and ahead of Apptronik on cumulative dollars, but below Figure AI's reported totals and far below the implicit valuation of Tesla's Optimus program.
The humanoid robot market in 2026 is crowded, and Digit is one of perhaps a dozen credible contenders. The table below shows publicly reported specifications and deployment status for the main commercial humanoids that compete with Digit for warehouse and factory work. Figures should be treated as approximate; many vendors revise these numbers between iterations.
| Robot | Maker | Height | Mass | Payload | Runtime | Commercial status |
|---|---|---|---|---|---|---|
| Digit | Agility Robotics | ~1.75 m | ~65 kg | ~16 kg | ~4 hours | Paying RaaS contract at GXO; 100k+ totes moved |
| Apollo | Apptronik | ~1.73 m | ~73 kg | ~25 kg | ~4 hours with hot-swap batteries | Pilots at Mercedes-Benz, Jabil, GXO |
| Figure 03 | Figure AI | ~1.68 m | ~60 kg | ~20 kg | ~5 hours | Fleet operations at BMW Spartanburg |
| Optimus Gen 3 | Tesla | ~1.73 m | ~57 kg | ~20 kg | not disclosed | Internal use at Tesla factories |
| Neo | 1X Technologies | ~1.65 m | ~30 kg | ~20 kg | ~2 to 4 hours | Home pilots; not yet at warehouse scale |
| H1 | Unitree | ~1.80 m | ~47 kg | not disclosed | ~2 hours | Research and demonstration |
A few patterns stand out. Digit's payload is on the low end, since the robot was designed around plastic totes rather than heavier industrial parts. The key differentiator is operational maturity: Digit is the only humanoid as of early 2026 with a published cumulative throughput number from a paying customer and the only one running a formal RaaS contract at a contract-logistics customer. Where competitors lead is dexterity and form factor. Figure 03 has five-fingered hands. Apollo emphasizes higher payload and swappable batteries. Optimus is integrated with Tesla's vertical stack. Neo targets the home rather than the warehouse.
Digit has received generally favorable coverage in technical trade press, with some skepticism about the broader humanoid hype cycle.
IEEE Spectrum has covered most major Digit revisions in detail and has been broadly positive about Agility's engineering culture, while flagging that the commercial pace of humanoid robots is much slower than the marketing rhetoric suggests. The Robot Report and Robotics 24/7 have used Digit's commercial milestones as reference points when comparing humanoid programs. TIME named Agility Robotics to its 2024 Best Inventions list and put Digit on the cover, and named CEO Peggy Johnson to its TIME100 AI list in 2025.
The most consistent line of criticism has been about cost and speed. Digit's published price is around $250,000 per unit when sold rather than rented, which is steep relative to the median annual cost of a warehouse worker. The RaaS structure helps because it spreads the cost over time and bundles in support, but the unit economics of paying for a humanoid versus paying a human only pencil out in narrow categories. Some industry analysts have also pointed out that Digit's productivity per hour, while improving, is still lower than a trained human worker on the same tote-moving task, and that the value proposition depends on extended uptime, multi-shift coverage, and the ability to absorb dull repetitive work in chronically understaffed facilities.
A second thread of criticism is about pace of capability growth. Humanoid robots from competitors using more aggressive end-to-end learning policies have shown more eye-catching manipulation demonstrations in the past year. Digit's measured, model-based approach trades a slower capability curve for higher reliability in deployed work, and reasonable people disagree about which strategy will win at scale.
On the labor side, some commentary has framed humanoid deployments at Amazon and GXO as a step toward replacing human warehouse workers. Agility's public position is that Digit is targeted at jobs that are already hard to staff and that the deployment model is augmentation rather than wholesale replacement, but the broader question of warehouse automation and employment remains contested.