Agility's New Humanoid Robot Kneels Down to Avoid Hurting People

Agility Robotics has introduced a significant update to its Digit humanoid robot: when it detects that a collision with a human coworker is imminent and unavoidable, the robot will stop moving and squat down to lower its centre of gravity and reduce the risk of injury. It sounds like a small thing. It is actually a thoughtful piece of engineering that reflects how seriously the industry is starting to take the question of what happens when robots and humans share the same physical space.

Digit is already operating in commercial warehouses. Amazon has been running trials with the robot for well over a year, using it for tasks like moving totes and handling bin transfers. Agility has positioned Digit as a general-purpose humanoid designed for environments built around human movement, which is a more interesting design bet than purpose-built industrial arms or wheeled logistics robots that need modified environments to function.

The Safety Problem Is Harder Than It Looks

Robots in factories have historically been kept away from humans. The entire design philosophy of traditional industrial robotics was built around the idea that robots are fast, powerful, and inherently dangerous, so you put them behind cages and interlocks and keep people well away. That works fine when you can design the environment around the robot.

But the whole promise of a humanoid robot is that it can operate in spaces designed for people, which means it needs to deal with people being around. That creates a genuinely difficult engineering problem. A humanoid robot moving at useful working speeds carries significant kinetic energy. If it collides with a person at full speed, the consequences could be serious.

The squatting response Agility has implemented works on a few levels. First, lowering the centre of gravity reduces the robot's effective mass in a collision, because less of the robot's momentum is concentrated at head or torso height where the most vulnerable parts of a human body are. Second, stopping movement immediately means the robot is not adding further kinetic energy to the situation after detecting a problem. Third, the predictability of the response matters: if workers know that Digit will always stop and squat when it detects a potential collision, they can trust the system in a way that allows them to work alongside it naturally rather than being constantly anxious about it.

The Broader Context: Humanoids Are Moving Fast

Agility's update comes at a moment when the humanoid robot industry has moved from research curiosity to commercial reality faster than most people anticipated. Boston Dynamics, Figure AI, 1X, Tesla with Optimus, and a growing list of well-funded startups are all racing to put humanoids into working environments. The numbers being discussed for deployment are significant: Tesla has talked about deploying tens of thousands of Optimus units, and the broader industry has attracted billions in investment over the past two years.

But most of those deployments are still in controlled settings with limited human interaction. The harder challenge is dense human-robot collaboration, where the robot and person are doing complementary tasks in close proximity throughout a working day. That is where the safety question moves from theoretical to immediate.

Regulators have not yet produced comprehensive standards for humanoid robots in workplaces. OSHA in the US and equivalent bodies in Europe are still developing frameworks. In the absence of clear regulation, companies like Agility are essentially setting their own safety standards, which puts significant responsibility on them to get it right.

Why This Matters Beyond Warehouses

If humanoid robots are going to eventually work in hospitals, care homes, retail environments, and homes, the safety behaviours they develop now in warehouse settings will form the foundation of what comes later. Getting the baseline right, teaching a robot to respond to humans in a way that is predictable, readable, and genuinely protective of human wellbeing, matters enormously for long-term trust.

A robot that kneels when it might hurt you is a small thing. A robot industry that takes that instinct seriously is a much bigger thing.