Viral videos of robots falling or moving chaotically are not just amusing content. They are a wake-up call for the entire industry. Manufacturers of humanoid machines face a fundamental challenge: how to deploy a robot in a warehouse or office without creating a threat to humans. The industry is already responding by implementing multi-layered safety systems, from specialized Nvidia chips to a radical shift away from bipedal designs in favor of wheeled platforms.
Incidents involving robots, including severe injuries and even fatalities, are forcing developers to rethink safety architecture. The key problem lies in the very nature of humanoid machines. Unlike traditional industrial robots, which are rigidly programmed and predictable (deterministic systems), humanoids operate based on AI. They belong to probabilistic systems, where every action is a statistical estimate rather than a rigid command. This adds an element of unpredictability that requires a completely different approach to safety.
Technical Solutions: From Chips to Design
Nvidia has already introduced a specialized safety system based on Blackwell chips. The company's Senior Director of Robotics, Amit Goel, emphasizes that the safety system and functional system now work in close conjunction, rather than in isolation. This allows the robot to interpret sensor data and stop at the slightest threat—not just by detecting an obstacle, but by assessing the context.
Fort Robotics goes even further, developing controllers that gather information from multiple sources. This is not just about visually detecting a person, but about complex analysis: where they are, in what posture, and how reliable this information is. This is critical for the robot to make correct decisions in a dynamic environment.
Some manufacturers have taken the path of radical simplification. Dexmate creates robots on wheeled platforms with long manipulators. Co-founder Yuzhe Qin explains that the battery and electronics are housed in the platform, providing a low center of gravity and eliminating the risk of falling. A similar approach is taken by Cobot, whose wheeled robots move at walking speed and lack excessive power. Their CEO Brad Porter reasonably notes: "We're not trying to crush watermelons."
Standards and the Future
The issue of stability loss is so critical that the International Organization for Standardization (ISO) has created a separate expert group. New requirements are not expected until mid-2028. Meanwhile, with no unified rules in place, manufacturers are developing their own scenarios. German company Neura Robotics, for example, designs its 80-kilogram robot 4NE1 so that if a joint fails, it falls "folding like a building," minimizing damage.
My expert opinion: The market is at a turning point. We are witnessing not just a race for functionality, but a mature approach to risks. Abandoning legs in favor of wheels is not a step backward, but a pragmatic solution for specific tasks. Investors should closely monitor companies that embed safety at the architectural level, rather than as a "patch." Those who solve the problem of falling and AI unpredictability will gain a real competitive advantage in 2025-2026.