Viral videos of robots kicking children or performing chaotic dances are not just amusing curiosities. For the industry, this is a stark alarm signal: how to deploy a machine in a warehouse, factory, or hospital without injuring a person? The industry's answer is a multi-layered safety architecture, ranging from specialized Nvidia chips to a fundamental rejection of bipedal designs in favor of wheels.
From Determinism to Probability: A New Risk Paradigm
The key problem lies in the very nature of humanoid robots. Unlike traditional industrial manipulators that operate on rigid algorithms, these machines are controlled by artificial intelligence. They are probabilistic systems that make decisions based on statistical models. This makes them flexible but also unpredictable. As one functional safety expert aptly noted: "If a robot loses power, it could simply fall and crush you." This is no joke—the industry is already recording cases of severe injuries and even fatalities.
Multi-Layered Protection: From Chips to Architecture
Nvidia has already proposed a hardware solution: a safety system based on Blackwell chips, capable of interpreting sensor data and stopping the robot in hazardous conditions. Fort Robotics goes further, creating controllers that gather information not only from the robot's cameras but also from the surrounding infrastructure. This is not just about detecting a person in the work zone, but about complex analysis: where they are, in what posture, and whether the data can be trusted enough for the robot to make a critical decision.
Standards and Alternatives: 2028 as a Milestone
Currently, there are no unified international safety standards for humanoids. An ISO expert group is working on them, but publication is not expected until mid-2028. Manufacturers are forced to act at their own risk. German company Neura Robotics developed a "controlled collapse" scenario for its 80-kilogram robot 4NE1: in the event of a joint failure, the machine folds downward, minimizing injuries.
Other companies have taken a more radical path. Startup Dexmate simply abandoned legs, placing robots on stable wheeled platforms with a low center of gravity. Cobot went even further, creating machines that move at pedestrian speed and lack excessive gripping force. "We're not trying to crush watermelons," their founder quips.
My analysis: While the industry awaits standards, we are witnessing a classic conflict between innovation and safety. Abandoning legs is a pragmatic but temporary solution. The true breakthrough will come when AI systems learn not just to stop, but to anticipate and avoid danger at the level of human intuition. Until then, every new video of a falling robot will set the industry back a step in the public eye.