Humanoid robots are being built to work in places designed for people. That ambition creates a hard safety problem: a machine shaped for human spaces may also bring risks that older robot rules were not written to handle.
The clearest issue is physical. A humanoid robot can be large, heavy and mobile, and it may need power just to stay upright. But the standards debate is also moving beyond falls and collisions, into communication, trust, privacy, security and the psychological effects of putting human-like machines near workers, patients, older adults, children and the public.
Why balance is the first problem
A warehouse robot named Digit shows why the topic is urgent. Digit, made by Agility Robotics, has been used to handle boxes of Spanx, moving loads between trolleys and conveyor belts. It can lift boxes up to 16 kilograms, taking over some of the heavier work from human colleagues.
For now, Digit works in a restricted, defined area. Physical panels or laser barriers separate it from human workers. That separation matters because Digit, despite usually staying steady on its robot legs, sometimes falls.
At a trade show in March, Digit appeared to be moving boxes successfully before it suddenly collapsed, face-planting on the concrete floor and dropping the container it was carrying. The source article points out the obvious danger: no one wants a 1.8-meter-tall, 65-kilogram machine falling onto them.
Pras Velagapudi, chief technology officer of Agility Robotics, gave a simple example of how serious contact could be. He said the throat is vulnerable, and that even a fraction of the force needed to carry a 50-pound tote could seriously injure a person.
Why an emergency stop is not enough
Physical stability is the No. 1 safety concern identified by the IEEE Humanoid Study Group, which is exploring new standards for humanoid robots. The group argues that humanoids differ from industrial arms and existing mobile robots in ways that call for their own safety approach.
One key difference is that humanoids can be dynamically stable. Aaron Prather, a director at the standards organization ASTM International and the IEEE group’s chair, explains that these robots need power to stay upright. They use their legs, or other limbs, to balance.
That makes the familiar red emergency button less straightforward. In many traditional robots, cutting power stops motion. With a humanoid, removing power may make the robot fall, which could create a new hazard instead of ending one.
Agility Robotics is working on features for the latest version of Digit that address this problem differently. Instead of instantly depowering, the robot could slow down more gently when a person gets too close. It might put down whatever it is carrying and move to its hands and knees before powering down.
Federico Vicentini, head of product safety at Boston Dynamics, is chairing an International Organization for Standardization working group on a new standard for industrial robots that need active control to maintain stability. His view is that standards should define the safety outcome without dictating every engineering method. As he puts it, the goal should be standardized, while the solution remains up to the designer.
Defining the robot may be harder than it sounds
Before standards can be written, standards groups need to decide what they are actually covering. The word humanoid can sound obvious, but it raises practical questions. Does the robot need legs? Arms? A head?
Prather’s group suggests that standards may need to move away from the term entirely. Instead of judging machines by appearance, the group recommends a classification system based on capabilities, behavior and intended use cases.
That shift matters because the safety challenge is not limited to robots that look like people. The ISO standard Vicentini is working on refers to industrial mobile robots “with actively controlled stability.” That could include Boston Dynamics’ dog-like quadruped Spot, its bipedal humanoid Atlas, or robots using wheels or another mobility system.
This broader framing could help standards cover the real source of risk: machines that move through human environments while actively managing their own balance.
Humans need to understand what robots are doing
Physical safety is only part of the problem. If humanoid robots share space with people, they need to signal what they are about to do. People should not be surprised when a robot steps into an aisle or changes direction.
Digit already uses lights to show its status and direction of travel, according to Velagapudi. But more advanced indicators may be needed if robots are to work cooperatively, and eventually collaboratively, with people.
Voice is one option, but the source article notes its limits. A loud industrial setting may make audio hard to use. If several robots are operating in the same space, it may also be unclear which machine is trying to communicate.
The IEEE group’s recommendations include standardizing some visual and auditory cues, enabling a human override, and aligning a robot’s appearance with its actual capabilities. Those cues would help workers, end users and members of the public understand a robot’s behavior before they have to react to it.
Human-like design changes expectations
Humanoid robots also create a psychological safety challenge. People tend to anthropomorphize machines that look like them. That can make users overestimate what a robot can do, relax their guard, or become frustrated when the machine does not meet human-like expectations.
This issue becomes more sensitive when robots are considered for hospitals, elderly care environments, homes or roles involving emotional labor. The IEEE report recommends emotional safety assessments and policies that mitigate psychological stress or alienation.
Greta Hilburn, a user-centered designer at the US Defense Acquisition University, surveyed non-engineers about humanoid robots. People overwhelmingly wanted capabilities such as facial expressions, reading micro-expressions, gestures, voice and haptics. Her conclusion was blunt: people wanted everything, including something that does not exist.
That gap between expectation and reality could matter most around vulnerable populations. Hilburn notes that if a robot is not programmed to speak in a way that makes a human feel safe, the outcome could differ for a child or an older adult.
Prather and Hilburn also emphasize inclusivity and adaptability. Standards may need to address whether a robot can communicate with deaf or blind people, wait longer for people who need more time to respond, and understand different accents.
The larger point is simple. Humanoid robots may eventually leave restricted warehouse spaces and enter places where people are not trained to work around them. Before that happens, manufacturers, regulators and end users need a shared minimum bar for safety, communication and trust.