Digit 5: Planned Safety Systems and Deployment Evidence

Updated October 11, 2026: added the October 1 Agility–FORT agreement and clarified the distinction between planned Digit 5 features and predecessor experience, including the workcell boundaries in Agility’s deployment process. Originally published September 17, 2026.

Development status: Agility's current Digit 5 product page says the robot is in development, some safety features remain in development, and specifications are preliminary estimates from pre-production testing. Designs and capabilities can change; safety features do not eliminate all operational risk. These industrial announcements do not establish safety for household use.

October 1 update: safety beyond the robot

In their October 1, 2026 announcement, Agility and FORT said they signed a memorandum of understanding to expand their collaboration. Their stated architecture spans a safety pendant, on-robot communications and off-robot interfaces connecting Digit to external safety systems. The proposed Offboard Safety Bridge extends the intended integration beyond onboard sensing.

FORT's matching announcement corroborates the partnership statement; it is not independent testing. The release identifies definitive agreements, bridge development and expected benefits as forward-looking matters. It does not establish completed rollout or certification of the combined system.

The release attributes more than 65,000 operating hours to Digit's predecessors. Those hours must not be presented as Digit 5 runtime or as validation of its proposed bridge.

For an operator assessing this proposal, useful questions include:

• Which external systems and hardware/software versions are supported?
• What happens after communication loss, and what evidence documents that behavior?
• Who can use the manual override during setup, maintenance and unexpected events?
• Which exact configuration, tasks and operating conditions have been evaluated, by whom, and with what documented limitations?

These are requests for evidence, not capabilities established by the announcement. A partnership description is not a site acceptance test.

What Digit 5 actually is

Agility positions Digit 5 for logistics and warehouse work. Its current product page lists a preliminary 23 kg/50 lb payload, 2.2 m/7.2 ft reach and fleet integration plans. Those are proposed product specifications, not measured results from a deployed Digit 5 fleet. Agility's earlier Digit versions have operated at customer sites; the October 1 release names Schaeffler, GXO and Toyota Motor Manufacturing Canada as predecessor deployments. These references must not be read as confirmation of Digit 5 operation at those sites.

Agility’s September 15 launch announcement sets early access for the first half of 2027 and general availability by the end of that year. Those are manufacturer targets. Neither an order pipeline nor a planned market launch establishes when an individual installation will be ready.

What changed in Digit 5

Four upgrades matter for anyone tracking home humanoids.

1. Planned cooperative safety architecture. Agility describes human detection, motion cues and an independent safety controller, with responses including avoidance, stopping or sitting. These are proposed system behaviors; the product page states that safety features remain in development.

2. Earlier workcell inspections. Agility’s deployment account describes field inspections of workcells containing earlier Digit robots by an OSHA-recognized Nationally Recognized Testing Laboratory. That is a claim about a particular deployment context, not an OSHA certification of every Digit configuration. It does not establish validation of Digit 5 or its planned external interfaces.

3. Standards participation. The launch release identifies work on ANSI/A3 TR R15.108 and ISO 25785-1. Participation in standards development is not certification of a product or evidence of suitability for homes.

4. Planned operational improvements. Agility's product page lists a 10:1 run-to-charge ratio and 20 hours of work per 24 hours, explicitly qualified as preliminary estimates from pre-production hardware and software. Planned workflow expansion and swappable grippers are design intentions. These claims are not demonstrated sustained output at a customer's facility.

The compute partnership: NVIDIA Halos and IGX Thor

The September announcement names Agility as NVIDIA’s first Halos for Robotics launch partner and describes planned use of IGX Thor and Halos Core. These names identify proposed technology relationships, not a completed safety assessment.

For a buyer, the useful question is how the components work together in the offered configuration and what evidence supports the protective functions. A supplier should explain which functions depend on its main AI model, which have separate supervision, and what limitations apply. This is a request for an account of the design, not a prescribed architecture for every household robot.

Why does removing the safety cage matter for home robots?

Whether an industrial installation uses barriers depends on its tasks, equipment and assessed risks. Household settings raise different questions, including moving children, pets and people with different mobility needs. Evidence from a bounded workcell cannot settle those questions automatically.

Digit 5’s proposed functions suggest questions to investigate for a household product; they are not a household safety checklist or approval:

• Does the robot detect people in all directions, including children and pets at low height?
• What protective response is specified for uncertainty in this configuration, and what evidence supports that response for the task and surroundings?
• Can bystanders understand what it will do next from lights, sounds, or posture?
• Is there an independent safety supervisor separate from the main AI planner?
• Has any third party evaluated it against a published standard, not just the manufacturer's own test?

These questions require evidence for the specific robot and setting. The Digit 5 announcement does not demonstrate that it has cleared them, and industrial results cannot establish safety around children or pets.

What this means for home timelines

Digit 5 is positioned for industrial deployment, with the stated availability targets aimed at business customers. The sources reviewed here do not establish a household offer or evaluated home capabilities. Do not read the industrial roadmap as a home launch.

What it does change is the timeline logic. Home humanoids from companies like 1X NEO, Figure 03, and Unitree G1 raise deployment questions that also arise in industrial work: applicable product and installation requirements, charger logistics, aisle and doorway clearances, workflow design, and documented protective behavior around people. The appropriate requirements depend on the product and intended use; different marks, standards and assessments should not be treated as interchangeable approvals.

For a home humanoid, ask which evaluations cover the offered configuration and intended household setting. Standards participation and fleet hours provide context; neither replaces that answer. A manufacturer can sell a product while leaving important evidence gaps, so distinguish the commercial offer from the confidence you place in its safety claims.

Why the operating setting belongs beside a demonstration

The practical gap is between performing an action and providing a repeatable service. A demonstration can show a transfer while leaving its preparation, supervision and recovery work outside the frame. An operator needs those details to decide whether the proposed system fits an existing process. That is not a reason to dismiss demonstrations; it is a reason to identify what they actually show before drawing a broader conclusion.

Published operating records can help readers assess deployment experience. Videos alone neither disclose the full process nor establish whether other private testing has occurred. If the evidence is unavailable, the appropriate conclusion is that the reader cannot evaluate it, not that no engineering or learning took place.

What the earlier deployment process actually demonstrates

Agility’s August 4 account of its deployment process adds an important boundary to the operating-hour figures: it says fourth-generation Digit remains in a sectioned-off workcell during both its customer program and ongoing RaaS deployments. Its process starts with a defined workflow, moves to an on-site pilot and then measures operational impact. The company describes a 90-day operational dataset at the end of that program. This is the supplier’s description of its process, not a dataset published here for independent analysis.

That boundary changes the inference a reader can draw. Experience moving goods inside a configured workcell can inform a successor’s design, but it does not show what happens when that successor shares an unrestricted route with people. The article’s references to earlier deployments should be read with that limitation. Conversely, operating within boundaries is not evidence that the trial was meaningless. It lets an operator evaluate a specific job without pretending the evaluation covered every possible setting.

Define the job before discussing the fleet

Consider a hypothetical transfer between two workstations. A useful pilot brief would identify the starting container, permitted load, pickup position, destination and completion signal. It would also describe who supplies the next container and who resolves an exception. Those details distinguish a robot moving an object from a workflow that the customer can actually use. This is an example of an evaluation brief, not a description of a Digit customer’s results.

Request separate records for a successful transfer, a stopped attempt, a human-assisted recovery and a job left unfinished. Ask what time the throughput calculation includes: loading, waiting, charging, resets and service can be inside or outside a reported interval. Without that definition, two throughput figures may describe different portions of the work. No numerical target for Digit 5 can be derived from this example.

A change log is useful alongside the results. If a demonstration required a new container, a cleared aisle or a revised pickup height, record the change. The question is not whether any preparation occurred; it is whether the preparation is practical in the intended operation and included in the cost and staffing plan. A reader should be able to tell what the robot did and what the site did to make the task possible.

Ask who owns each part of the proposed safety connection

The Offboard Safety Bridge proposal creates an interface question as well as a robot question. An operator evaluating that proposal should request a diagram showing the connected equipment, the signals exchanged and the party responsible for each boundary. The announcement does not supply that site-specific design. A component name alone cannot answer whether an existing installation can use it unchanged.

Ask the supplier and qualified site team which documented conditions permit operation, which require a stop, and what checks are necessary before work resumes. Request the relevant configuration and assessment records instead of inventing a test involving an approaching person. These are documentation questions, not instructions for commissioning safety equipment or proof that a particular protective response has been implemented.

Responsibility after handover matters too. Identify who can approve software or hardware changes, who records them and who decides whether the evaluated setup still applies. A result obtained with one gripper, load or software version should remain attached to that configuration. This prevents a later feature announcement from quietly becoming evidence for a different system.

Keep service obligations beside performance results

A deployment proposal should name the support contact and distinguish response time from repair completion. Ask which parts, labor, transport and temporary replacement arrangements are included, and which require a separate agreement. If a charging or service interruption occurs, who owns the unfinished work? These are questions to put in a procurement discussion; the linked announcements do not establish their answers for a particular customer.

The same reasoning is useful at home without importing warehouse approval into a living room. A household should know what assistance, setup and support a vendor expects from it. A successful industrial pilot can demonstrate something valuable within its stated conditions. It cannot substitute for evidence about a different robot, task, environment or service contract.

How Digit 5 compares to the home humanoid contenders

Nobody should cross-shop an enterprise RaaS biped against a consumer preorder. But the comparison is useful as a readiness audit — it shows which home contenders have deployment proof and which have promises.

The pattern: price and availability tell you what you can buy; safety evidence tells you what you should trust. These sources do not establish a comparable safety ranking across the robots. Predecessor experience and proposed features need separate treatment.

The ui44 verdict: trust the safety evidence

Digit 5 does not put a humanoid in your home. It puts a filter on every company that promises to. The next time a viral video shows a humanoid folding laundry beside a child, ask the Digit 5 questions: who evaluated its safety, what standard did they use, what does it do when uncertain, and how many hours has it worked beside real people? Until those answers exist, keep your wallet closed and keep watching the warehouses — that is where home readiness is being earned.

Browse every platform we track, from warehouse bipeds to home companions, in the Home Robot Database, compare contenders side by side in Compare, and follow the category in Humanoids.

That is the Digit 5 lesson for home buyers: trust the safety evidence, not the sizzle reel.