Agility Digit 5 Humanoid Robot Drops the Safety Cage, but the Safety Case Is Still Being Built

Agility Robotics unveiled the Agility Digit 5 humanoid robot with one consequential promise: it can work near people without the physical barriers surrounding earlier models. The company says new sensors, AI-based human detection, and an independent safety controller let Digit avoid, stop, or sit when someone approaches.

That change matters because Digit 4 has already moved beyond staged demonstrations. Agility says those robots have logged more than 65,000 operating hours across customer facilities in North America. Yet deployments have largely kept the machines inside protected workcells, limiting how freely they can move through facilities designed around people.

Digit 5 attempts to remove that constraint while carrying heavier loads and handling more varied tasks. It also arrives as Agility prepares to enter public markets through a planned transaction valuing the company at $2.5 billion. The central test is no longer whether a humanoid can move a tote. It is whether safety software can replace a physical boundary without transferring unacceptable risk to nearby workers.

Agility Digit 5 Moves Beyond the Robot Workcell

Digit 5 changes Agility’s operating model by treating human proximity as a condition to manage, not a reason to fence off the robot.

Agility introduced Digit 5 on September 15, 2026, as the fifth generation of its general-purpose humanoid. Early access is expected during the first half of 2027. General availability for manufacturing, warehouse, and distribution operators is planned by the end of that year.

The defining change is cooperative safety, meaning the robot is designed to operate near people while adjusting its behavior as conditions change. Earlier Digit deployments used workcells, which are controlled areas separated from workers by barriers.

A workcell offers a simple safety principle. If people cannot enter the robot’s operating space, they face fewer hazards from its motion, payload, or loss of balance. However, that boundary also limits the robot’s ability to move between stations and participate in workflows that require frequent human interaction.

Digit 5 replaces that fixed separation with a layered response. According to the company’s Digit 5 announcement, its onboard systems monitor for people and choose an action based on their distance and movement.

At longer distances, the robot can alter its route or stop and wait. If someone moves closer, Digit can lower itself into a seated position. Visual and audible signals are intended to tell workers what the robot is doing before it moves again.

The robot does not rely solely on its primary AI system to make those decisions. Agility says an independent safety controller monitors unsafe proximity and triggers the appropriate response. That separation matters because a safety system should remain capable of intervening when the main controller encounters a software fault.

Digit 5 also processes visual and movement data locally. Its equipment includes onboard cameras, depth sensors, and inertial measurement units, which measure the robot’s orientation and acceleration. Agility says only system health and diagnostic data must travel to its Arc fleet-management platform.

The barrierless design expands the workflows Agility can pursue. A robot confined to one cell can repeatedly move containers between fixed points. A robot that shares aisles with workers can potentially connect several stages of a warehouse or manufacturing process.

Agility identifies depalletizing, machine tending, kitting, sequencing, quality inspection, and palletizing as target activities. Some remain planned capabilities rather than independently verified production functions. The distinction is important because the hardware is still in development.

The company’s published specifications are preliminary. It also warns that certain safety features remain under development and that no safety system eliminates every operational risk.

Those qualifications make Digit 5 a product launch and a safety proposition. Agility has disclosed the architecture, intended behaviors, and release window. Customers still need to see how that architecture performs through long shifts, crowded aisles, changing lighting, and unpredictable human movement.

Why Removing Barriers Changes the Business Case

A humanoid cannot automate a connected facility workflow if every task requires its own isolated operating zone.

Digit 4 has primarily handled tote movement. At GXO’s facility near Atlanta, Agility says the robot crossed a cumulative milestone of 100,000 totes while maintaining approximately 98% accuracy when actively performing its assigned work.

That record gives Digit 5 a more substantial foundation than a laboratory prototype. Agility reports nearly three years of commercial deployment involving GXO, Schaeffler, Amazon, and Toyota Motor Manufacturing Canada. Still, operating hours do not automatically validate every new safety behavior introduced with Digit 5.

The more immediate lesson concerns workflow design. Warehouses contain workers, forklifts, autonomous mobile robots, conveyors, doors, staging areas, and equipment that cannot always be rearranged for a single machine. Adding a fenced workcell can consume space while creating new handoff points.

Consider palletizing. A humanoid might place packages on a pallet, but a person must eventually remove or replace that pallet. If the entire area requires physical isolation, each handoff creates delay and additional control logic.

A barrierless robot can pause while the worker completes that exchange. It can then resume without forcing the facility to build another permanent enclosure. Agility Chief Technology Officer Pras Velagapudi told Ars Technica that such scenarios make a physical cage impractical.

The same logic applies to machine tending. A mobile robot might carry parts between machines, load them, and collect finished items. That sequence becomes difficult when safety barriers divide every movement into isolated segments.

This is the practical meaning behind the Digit 5 warehouse impact. Its value depends less on resembling a person than on navigating human-scale spaces and linking tasks that existing automation handles separately.

Agility has redesigned the hardware for that broader role. Digit 5 can repeatedly lift loads of up to 50 pounds, a 40% payload increase over its predecessor. It stands 5 feet 11 inches and reaches up to 7.2 feet, compared with Digit 4’s 5.5-foot reach.

The robot weighs 284 pounds. That mass explains why human detection and controlled stopping carry such importance. A software mistake involving a tabletop device is inconvenient. A mistake involving a moving industrial machine can cause serious injury.

Digit 5 also uses interchangeable grippers attached through standardized mounting flanges. These end-effectors, meaning tools placed at the end of a robotic arm, let one machine use different hardware for tote handling or other manipulation tasks.

Battery performance addresses another commercial constraint. Agility claims a 90-minute runtime followed by a nine-minute charge, creating a 10-to-1 operating-to-charging ratio. Digit 4’s ratio was 2-to-1.

Agility estimates that this schedule can support more than 20 productive hours within a 24-hour period. That figure comes from internal testing and benchmarking of preproduction systems. Real facilities will need to verify it under their own payloads, temperatures, routes, and charging schedules.

The business argument therefore combines utilization and flexibility. More time working improves the return on each deployed machine. More reachable tasks let operators spread integration costs across a larger workflow.

Agility says it had more than $300 million in multi-year Digit 5 orders as of May 2026. A related SEC investor filing says those orders concern 1,000 robots under a three-year service contract.

Those orders are not current-period revenue. They remain subject to contractual milestones, and associated warrants vest as robots are deployed. That makes successful delivery and production performance more important than the headline amount alone.

How Agility Digit 5 Works Around People

The safety mechanism combines perception, controlled behavior, redundant supervision, and communication with nearby workers.

The first layer is human detection. Digit 5 uses proprietary AI algorithms and several vision-based sensors to monitor its surroundings. Agility has not publicly specified every sensor type or the detection ranges used for different responses.

AI helps interpret sensor data, but interpretation alone is not a safety case. The robot must connect detection to predictable behavior within a known amount of time. It must also respond conservatively when readings conflict or visibility deteriorates.

Digit’s response can vary with proximity. At a safe distance, it might choose another route. If a person crosses its path, it can stop and allow that person to pass. At close range, it can squat to reduce the danger associated with an upright, dynamically balancing machine.

That graduated behavior differs from a single emergency shutdown. An immediate power cut can be necessary, but it is not always the safest response for a legged robot carrying a load. Controlled motion can place the machine and its payload into a more stable state.

An independent controller adds another layer. It supervises the distance between Digit and nearby people, then activates a safety response when required. This design separates safety enforcement from the system selecting and executing ordinary tasks.

Agility is pairing its architecture with Nvidia IGX Thor, a computing platform for safety-sensitive robotic and medical applications. It is also an early launch partner for Nvidia’s Halos for Robotics software and safety infrastructure.

The hardware partnership gives Agility access to computing designed around functional safety. Functional safety focuses on whether automated protections respond correctly when a system detects a dangerous condition or component failure.

It does not mean that every AI-generated movement becomes safe by default. Agility must establish which components perform safety-rated functions, how failures are detected, and what state the robot enters after a fault.

The robot also communicates intent through visual and audible signals. A worker needs to distinguish a robot that has yielded from one preparing to resume movement. Clear cues become especially important when several machines occupy the same space.

This human factor is easy to underestimate. Workers cannot inspect an algorithm’s internal state while walking through a loading area. They rely on visible behavior, training, procedures, and consistent signals.

The design also depends on fleet-level integration. Agility Arc connects Digit with warehouse management, warehouse execution, and manufacturing execution systems. It tracks operating measures such as uptime, throughput, and mean time between incidents.

Those connections can help facilities coordinate handoffs and stop work when an external safety system detects a problem. They also create more interfaces that operators must configure, validate, and protect.

Agility says data sent to Arc uses encrypted connections, while customer information remains separated. Local processing reduces the amount of raw sensor data leaving the facility. Enterprise buyers will still need clear retention rules, access controls, and incident procedures.

Understanding how Agility Digit 5 works therefore requires looking beyond collision avoidance. The AI component detects and interprets people, but safety also depends on independent controls, stable fallback behavior, facility integration, and worker communication.

This layered approach reflects the limits of perception software. Cameras can encounter glare, shadows, dust, blocked views, unusual clothing, or partial visibility. Depth sensors can disagree with other inputs. A cautious design must handle uncertainty without improvising around it.

Traditional industrial systems often make their safe boundaries visible. A fence defines where people should not stand. Digit 5 replaces much of that physical clarity with sensing and dynamic separation, making system validation more demanding.

The Safety Claim Still Faces a Standards Gap

Digit 5’s strongest claim cannot be settled by a launch demonstration because the relevant humanoid safety standards are still developing.

Agility says Digit previously completed an independent field evaluation involving industrial safety standards administered under a program connected to the U.S. Occupational Safety and Health Administration. That achievement concerns an earlier deployment configuration, not blanket certification for every Digit 5 feature.

The company is participating in work on ANSI/A3 TR R15.108, a technical report for dynamically stable industrial mobile robots. It is also contributing to ISO 25785-1, an international standard under development for industrial mobile robots, including humanoids.

As of the Digit 5 announcement, ISO 25785-1 remained under committee review. According to robot safety reporting, the proposal must eventually proceed to a vote involving the standard’s participating national members.

This timing creates an unusual situation. Agility is designing a product for a category whose detailed safety framework is not yet complete. The company can draw on existing machinery and robotics requirements, but a dynamically balancing humanoid introduces distinct failure modes.

A wheeled mobile robot can stop without worrying about remaining upright. A humanoid must manage balance, limb position, carried objects, and the possibility of falling. Its height and mass make a fall relevant even when its arms stop moving.

Digit 5’s 284-pound weight sharpens that concern. Sitting down is a sensible mitigation only if the robot can execute the motion reliably and avoid creating another hazard. Payloads also need a controlled response during sudden stops.

The hardest test will involve edge cases rather than routine operation. Can the system detect a person emerging from behind a pallet? How does it respond when two workers approach from different directions? What happens after a sensor loses visibility during a turn?

Facilities must also manage predictable human adaptation. Workers may become comfortable walking closer to a robot after weeks without an incident. They may misread an alert in a noisy environment or assume that Digit has seen them.

No sensor array eliminates those behavioral risks. Training, marked pathways, speed restrictions, maintenance, and local hazard assessments will remain part of the deployment. Removing a cage does not remove the employer’s responsibility to control hazards.

Agility’s own product materials acknowledge the limit. Digit 5 remains in development, specifications can change, and some safety functions are unfinished. The company also states that the features do not eliminate all operational risk.

That language should guide how buyers interpret the launch. Digit 5 is engineered for barrierless operation, but broad commercial evidence will arrive after customers deploy the production model in 2027.

The distinction also affects competition. Tesla’s Optimus program attracts attention through its connection to automotive manufacturing and Elon Musk’s large production forecasts. Figure AI has promoted general-purpose workplace tasks, while Apptronik is developing Apollo for industrial environments.

Agility’s current advantage is practical deployment experience. The public-market announcement identified Amazon, Nvidia, SoftBank, and Foxconn among its backers, with Toyota and Schaeffler among early customers.

Yet experience with Digit 4 does not automatically transfer to cage-free Digit 5 operations. New sensors, controls, batteries, actuators, tasks, and human interactions create a larger validation surface.

Competitors face the same pressure. A humanoid that requires permanent separation may struggle to justify its form factor against fixed automation or wheeled robots. A machine that shares space with people must meet a much higher evidentiary threshold.

That is the central tradeoff. Removing barriers can unlock more flexible automation, but it makes sensing reliability and controlled behavior responsible for protections that workers could previously see.

Three Signals Will Show Whether Digit 5 Can Scale

Digit 5 becomes consequential only when customer deployments verify its safety, utilization, and workflow claims under production conditions.

The first signal is the quality of early-access deployments during the first half of 2027. Agility needs more than operating-hour totals. Buyers should look for incident rates, protective stops, intervention frequency, task completion, and performance around unscripted human movement.

A successful pilot would show workers and Digit sharing active areas without frequent shutdowns or improvised restrictions. If customers restore temporary fences or dedicate empty aisles to the robot, the practical benefit of cooperative safety will weaken.

The second signal is progress on standards and third-party evaluation. ISO 25785-1 and related North American work can clarify testing methods for dynamically stable industrial mobile robots.

Published assessments should explain the operating conditions they cover. A passing result for one task, speed, payload, or facility layout should not be treated as approval for every possible workflow.

Third-party evidence would strengthen Agility’s claim by separating product marketing from repeatable safety performance. Delays, narrow certifications, or extensive operating restrictions would show that barrierless deployment remains harder than the launch presentation suggests.

The third signal is conversion of the order book into functioning fleets. Agility’s more than $300 million in orders represents potential multi-year value, not recognized revenue or completed delivery.

Investors and customers should track how many robots enter service, how many remain active, and whether customers expand after initial installations. A large contracted fleet matters only when deployment milestones are met and customers accept the resulting performance.

Production capacity will shape that conversion. Digit is assembled at RoboFab, Agility’s 70,000-square-foot facility in Salem, Oregon. The company says the plant is designed for annual capacity of up to 10,000 robots and more than 500 employees when fully built out.

Designed capacity is not current output. Scaling a complex machine requires reliable suppliers, manufacturing quality, field service, spare parts, and technicians who can diagnose problems quickly.

Agility also faces scrutiny from its planned public listing. The proposed SPAC transaction values the company at $2.5 billion, increasing pressure to turn technical progress into measurable commercial results.

That pressure does not invalidate the technology, but it raises the cost of vague reporting. Fleet size, customer concentration, uptime, deployment duration, and milestone completion will offer a clearer picture than cumulative orders alone.

For enterprise buyers, the next year should be treated as a validation period. Teams evaluating humanoids will need to compare the robot’s output with the full cost of integration, supervision, charging, maintenance, and safety management.

They will also need organized records from trials, risk assessments, worker feedback, and incident reviews. A searchable engineering knowledge base can help teams preserve that evidence across operations, safety, and engineering groups.

The Agility Digit 5 humanoid robot gives the industry a concrete test instead of another choreographed demonstration. Its hardware has a release window, its target workflows are defined, and its safety architecture has identifiable components.

What remains unresolved is whether that system can manage ordinary disorder. Warehouses are noisy, crowded, and frequently reconfigured. People behave less predictably than machines, while equipment degrades between scheduled tests.

By the end of 2027, the useful question will not be whether Digit 5 can walk beside a person on camera. It will be whether customers trust it beside workers for thousands of hours, expand their fleets, and keep the barriers out of the building.