Digit 5 Humanoid Safety: What 'Truly Safe' Means

Digit 5 humanoid safety does not work the way most people assume. The design does not pad the arms or slow the joints. It removes the fall. IEEE Spectrum’s Evan Ackerman reported on September 15, 2026 that Agility Robotics aims to make it “physically impossible for Digit to fall over on anyone, ever,” by having the robot autonomously avoid, stop, or assume a seated position when a person approaches.

A robot that sits down when you walk towards it sounds almost comic. Do the arithmetic on a 129-kilogram machine standing 1.8 metres tall and it stops being funny.

Key Takeaways

• Published figures: 1.8 metres tall, 129 kilograms, a 23-kilogram payload lifted to 2.1 metres, and operation for more than 20 hours a day. The robot “can work in close proximity to people without relying on physical barriers.”
• The safety strategy is energy management by posture. Detect a person, put down whatever is being carried, then sit. That addresses the two largest energy stores on the machine.
• My own arithmetic from those published numbers: a 129 kg mass with its centre of mass roughly 0.9 m up carries on the order of 1,100 joules of gravitational potential energy. A 23 kg payload at 2.1 m carries roughly 470 joules.
• Ackerman’s own verdict is measured: the approach “may not be elegant, but it works,” while “there are still no easy answers” to humanoid safety.
• The claim is unproven in the field. The article notes Agility does not have Digit 5 deployed in factories yet, cites no ISO standards, no force or torque limits, and no third-party certification.

Digit 5 Humanoid Safety: The Numbers Agility Published

Start with what is on the record, because the record is short and specific.

Height: 1.8 metres. Mass: 129 kilograms. Payload: 23 kilograms, lifted to 2.1 metres. Duty cycle: more than 20 hours per day. And the central operational claim, that it can work close to people without physical barriers.

That last point is the commercial one. Caged robots need floor space, fencing and interlocks, and they cannot share a workspace with people. A machine that works barrier-free changes what a warehouse layout can look like, which is why “safe” here is a business requirement rather than a courtesy.

Now what the article does not contain, which is just as informative. No ISO standard is named. No collision force or torque limit is given. No sensor types are specified for human detection. No third-party certification is mentioned. No Agility engineer is quoted by name with a title.

That is not an accusation. It is a map of what a parent, a journalist or a safety manager still cannot check.

Why the Fall Is the Hazard, Not the Arm

Here is the mechanism, and it is the part that makes the design make sense.

Gravitational potential energy is mass times gravity times height, written mgh. Using the published mass of 129 kilograms, a centre of mass somewhere around 0.9 metres for a standing humanoid, and 9.81 metres per second squared, you get roughly 1,100 joules. That is my calculation from Agility’s published figures, not a number Agility released, and you can redo it in your head to check me.

Now the payload. A 23-kilogram object at 2.1 metres gives 23 × 9.81 × 2.1, which is about 470 joules.

Those two numbers explain the entire safety sequence. The biggest energy store on the machine is the machine itself, standing up. The second biggest is whatever it is holding, held high. So the mitigation is: put the object down, then lower the robot’s own centre of mass. Sitting is not a gesture. It is a reduction in stored energy.

Compare this to the usual approach. Collaborative robot safety traditionally focuses on contact: limiting speed, limiting force, detecting a touch and stopping. That works well for an arm bolted to a table, because the arm is the only thing that can hit you and the table is not going anywhere. A free-standing humanoid breaks that model, because the dominant hazard is no longer the end effector. It is 129 kilograms of inverted pendulum.

The analogy, now that the numbers are in place: a ladder is not dangerous because of its rungs. It is dangerous because it is tall and you are on it. Padding the rungs does nothing. Making the ladder sit down does.

Three Safety Strategies, and How Each One Fails

Strategy	How it works	Where it is used	How it fails
Separation	Cages, fences, light curtains, interlocks	Classic industrial robots	Fails during maintenance, when people are inside the cage
Speed and separation monitoring	Robot slows or stops as a person nears	Mixed workcells	Fails if sensing misses the person or misjudges distance
Power and force limiting	Robot physically cannot exert dangerous force	Collaborative arms	Fails for large mobile machines, where mass itself is the hazard
Posture and energy management	Robot lowers its own stored energy before contact is possible	Digit 5’s stated approach	Fails if human detection fails, or if the sit takes too long
Administrative controls	Training, procedures, exclusion zones	Everywhere, as a backstop	Fails under time pressure and on the night shift

The context for all of this is thinner than people expect. OSHA states plainly that “there are currently no specific OSHA standards for the robotics industry,” and directs users to its machine-guarding materials instead. It also notes that “many robot accidents occur during non-routine operating conditions, such as programming, maintenance, testing, setup, or adjustment.”

Consensus standards exist and are where the real detail lives. ISO 10218 covers industrial robot safety in two parts, robots and robot systems and integration, developed in 2011, and ISO/TS 15066 was published in 2016 to address collaborative robots specifically, a category the original standard did not anticipate. A free-standing bipedal humanoid working barrier-free among people is another category that existing documents were not written for, which is the honest reason no standard is cited.

How to Teach Your Kid About Robot Safety

Ages 5–8: The tower that sits down

Materials: wooden blocks or stacking cups, a clear floor.

Build two towers of the same blocks: one tall and thin, one short and wide, same number of pieces. Push each one over with the same gentle nudge. The tall one goes further, makes more noise and scatters more.

Ask why, and steer the answer toward height. Then the punchline: “if the tall tower could sit down when you walked up to it, would it still be dangerous?” That is Digit 5’s whole safety argument, and a six-year-old can now explain it.

For a second round, put a small toy on top of the tall tower and knock it over again. The toy flies further than anything else. That is the payload, and it is why the robot puts the box down first.

Ages 9–12: Measure height against damage

Materials: a baking tray, flour or fine sand levelled flat, a marble or small ball, a ruler, a phone to photograph results.

Drop the same ball from 20 cm, 50 cm and 100 cm into the levelled flour. Measure the crater depth each time and write the three numbers down. The relationship is not subtle.

Now the engineering question, which is the real exercise. Hand your child a budget of three safety features for a robot and ask where to spend them: softer hands, slower arms, or a mechanism that makes it shorter when a person approaches. Make them justify the answer using their own crater data. Most kids pick the height mechanism once they have the numbers, and that is the same conclusion Agility’s engineers reached.

Ages 13+: Write the safety case

Three parts, on one page, using the published Digit 5 figures.

First, the arithmetic: compute mgh for the robot and for the payload yourself. Show the working. Compare the two.

Second, the hazard list: what could hurt someone, ranked by energy rather than by how scary it looks. Falling, dropped payload, pinching, arm impact, tripping over the robot’s feet.

Third, and this is the part that matters, the evidence requirement: write down what would have to be true for you to believe the safety claim. Named standards. A certification body. Hours of barrier-free operation logged. A published incident rate. Then check which of those exist today. Currently, none of them are in the public record, and noticing that gap is the skill.

The question to ask: “What is the most dangerous thing about this robot that is not its arms?”

What to Do at Home

Read engineering claims for the mechanism, not the adjective

“Truly safe” is an adjective. “Sits down when a person approaches, after putting the payload on the floor” is a mechanism. Teaching a kid to hunt for the second kind of sentence in any product announcement is a transferable habit, and robotics announcements are unusually good practice because the mechanisms are physical and checkable.

Use the energy frame on household hazards

The same arithmetic explains why a toddler falling off a chair is less serious than falling down stairs, why a dropped glass breaks and a dropped sock does not, and why bike helmets matter more going downhill. Once a kid has mgh as an intuition rather than a formula, a lot of safety rules stop needing enforcement.

Separate “designed to be safe” from “proven safe”

The IEEE Spectrum piece is careful about this and parents should be too. Digit 5 is designed around a credible hazard analysis. Ackerman notes it is not deployed in factories yet. Design and proof are different stages, and conflating them is how both hype and panic get started.

Watch the maintenance case, because that is where people get hurt

OSHA’s observation that many robot accidents happen during programming, maintenance, testing, setup or adjustment is the single most useful fact in this article for anyone whose family member works near machines. The safety story a company tells is usually about normal operation. The injuries happen when something is being fixed.

What not to do

Do not let a kid conclude that humanoids are either safe or dangerous as a category. The useful question is always about a specific machine, a specific hazard and a specific mitigation, with a number attached. A child who learns to ask “how heavy, how tall, and what happens when it fails” is equipped for the next thirty years of this industry, whatever shape the robots end up having.

What to Watch For Over the Next 3 Months

• Week 4: Check whether Agility has published a safety conformance claim naming ISO 10218, ISO/TS 15066 or a successor document, or announced a third-party assessment. That would be the first hard external evidence beyond the design argument.
• Month 2 red flags: Marketing that keeps repeating “truly safe” without adding a standard, a force limit or a deployment count. Also watch for the opposite signal, which is good news: a disclosed incident with an explanation of what the sensing missed.
• Month 3 self-check: Ask your kid which part of the safety sequence they would attack if they wanted to break it. The right answer is human detection, because everything downstream depends on noticing the person in time. If they say “the arm,” run the block-tower activity again.

Frequently Asked Questions

Is Digit 5 actually safe?

It is designed around a defensible hazard analysis, and that is a different claim from being proven safe in service. IEEE Spectrum reported on September 15, 2026 that Agility does not yet have Digit 5 deployed in factories, and the coverage names no ISO standard, force limit or third-party certification. “Promising and unverified” is the accurate summary.

Why does sitting down make it safer?

Because stored energy depends on height. Using Agility’s published mass of 129 kilograms and a standing centre of mass near 0.9 metres, the gravitational potential energy is on the order of 1,100 joules. Lowering the centre of mass reduces that figure, and a seated machine cannot topple onto someone.

Why put the payload down first?

Same reason, applied to the heaviest thing it is holding. A 23-kilogram object at 2.1 metres carries roughly 470 joules by the same calculation. Releasing it to the floor removes the second-largest energy store before the robot moves.

Are there safety standards for humanoid robots?

Not specific ones, which is the honest answer. OSHA states there are currently no specific OSHA standards for the robotics industry. ISO 10218 addresses industrial robots and ISO/TS 15066 addresses collaborative robots, and neither was written with free-standing bipedal machines working barrier-free in mind.

Should this change how I feel about robots in warehouses?

It should change what you ask. The useful questions are about the hazard analysis, the sensing that triggers the safe behaviour, the number of barrier-free operating hours logged, and what happens during maintenance. Those are answerable. “Are robots safe” is not.

Is a 129-kilogram robot heavier than I expected?

Most people guess much lower, and the gap matters. For comparison with something familiar, that is heavier than most adults, standing at adult height, and designed to operate for more than 20 hours a day. Mass is the reason the safety engineering focuses where it does.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Ackerman, E. (2026, September 15). “Digit 5 May Be the First Humanoid Robot Worker That’s Truly Safe.” IEEE Spectrum. https://spectrum.ieee.org/humanoid-robot-safety
2. Occupational Safety and Health Administration. “Robotics.” US Department of Labor. https://www.osha.gov/robotics
3. Wikipedia contributors. “ISO 10218.” Wikipedia. https://en.wikipedia.org/wiki/ISO_10218
4. Wikipedia contributors. “Inverted pendulum.” Wikipedia. https://en.wikipedia.org/wiki/Inverted_pendulum
5. Wikipedia contributors. “Moravec’s paradox.” Wikipedia. https://en.wikipedia.org/wiki/Moravec%27s_paradox
6. Wikipedia contributors. “Humanoid robot.” Wikipedia. https://en.wikipedia.org/wiki/Humanoid_robot
7. OpenAI, Andrychowicz, M., Baker, B., et al. (2018). “Learning Dexterous In-Hand Manipulation.” https://arxiv.org/abs/1808.00177

Related reading on HiWave Makers: what factory robots do that home robots cannot, humanoid robots, kids, safety and attachment research, and why humanoid robots fall over.