Safety first: how humanoid robot manufacturers protect people from "iron" mistakes
Viral videos of robots kicking children or going out of control have become a wake-up call for the entire industry. The main question facing developers today is: how to deploy a machine in a warehouse, hospital, or factory without creating a threat to humans. The industry has already found several answers — from hardware solutions by Nvidia to completely abandoning bipedal designs in favor of wheeled platforms.
Incidents involving humanoid robots are not uncommon. Some companies have already reported severe injuries and even fatalities. This is forcing developers to radically rethink safety approaches, which undoubtedly slows down the mass adoption of such devices.
The Problem of Probabilistic Systems
The key challenge lies in the architecture. Traditional industrial robots (welders, loaders) are deterministic systems operating on rigid algorithms. Humanoid robots controlled by AI, on the other hand, belong to probabilistic systems. They make decisions based on statistical data rather than clear rules, making their behavior less predictable and requiring multi-layered protection.
Nvidia has already proposed its solution — a safety system for humanoid robots based on Blackwell chips. As explained by the company's Senior Director of Robotics, Amit Goel, this model can interpret sensor data in real time, identify potential hazards, and instantly stop the robot's operation in unsafe conditions. Essentially, a software stack has been created that allows the safety system and functional system to work in close coordination.
Another approach is to transfer part of the control to the infrastructure. Fort Robotics, based in Philadelphia, is developing controllers that collect information not only from the robot's own cameras but also from external sensors. CEO Samuel Reeves emphasizes that it is now important not just to detect a person in the work zone, but to understand their posture, location, and the degree of reliability of this data for decision-making.
How Not to Fall Flat on Your Face
The problem of loss of stability is so serious that an expert group from the International Organization for Standardization has taken up its study. The publication of unified requirements is expected only by mid-2028. In the meantime, manufacturers are seeking their own solutions.
Germany's Neura Robotics has equipped its 80-kilogram bipedal robot 4NE1 with a controlled fall system. If the machine detects a malfunction, such as a joint failure, it attempts to regain balance, and if that is impossible, it folds up like a "collapsing building," minimizing damage.
Some companies have decided to eliminate the risk of falling altogether. Dexmate creates robots on wheeled platforms with a low center of gravity, where the battery and electronics are placed in the base. As co-founder Yuzhe Qin notes, such a machine physically cannot fall. Cobot founder Brad Porter adds that their wheeled manipulators, which push carts in hospitals, move at walking speed and do not have excessive gripping force. "We're not trying to crush watermelons," he jokes.
Analytical commentary from Cryptalist: While the market awaits unified safety standards, the race for "human-likeness" may turn out to be a dead end. Wheeled platforms and control through external infrastructure appear to be a far more pragmatic and safe solution for industrial applications than attempts to make a bipedal robot dance without risking human lives.