Viral videos of robots kicking children or going out of control have highlighted the industry's main dilemma: how to deploy a humanoid machine in a warehouse without injuring employees. The industry is responding to this challenge with multi-layered safety systems — from specialized Nvidia chips to a fundamental rejection of legs in favor of wheeled platforms.
Various "incidents" involving robots are forcing developers to radically rethink safety mechanisms, which in turn slows down mass adoption. Several companies have already reported severe injuries and even fatalities related to humanoid devices. As robot functional safety expert Michelle Silva notes, the simplest threat is that when power is cut off, a humanoid robot can simply fall and crush a person.
The key complexity lies in the architecture. Traditional industrial robots — welding machines or loaders — are deterministic systems operating on a rigid set of rules. Humanoid robots controlled by AI, however, belong to probabilistic systems. They act based on statistical probabilities rather than absolute predictability, requiring fundamentally different approaches to safety when working near humans.
Multi-layered protection: from chips to infrastructure
Nvidia introduced a safety system for humanoid robots based on Blackwell chips. The company's Senior Director of Robotics, Amit Goel, explained that the model interprets sensor data about potential hazards and can emergency-stop the robot in unsafe conditions. This is not just an isolated function but a new level of the operating system, allowing safety systems and functional modules to run in a unified context.
Another level of control assumes that the robot will rely not only on its own cameras and sensors but also on external infrastructure. Fort Robotics is developing controllers that gather information from multiple sources — not just detecting a person's presence in the work zone but also analyzing their posture, position, and the reliability of this data for decision-making.
The issue of stability loss is so critical that it is being studied by an expert group of the International Organization for Standardization. The publication of unified requirements is not expected until mid-2028. In the meantime, manufacturers are developing their own scenarios. German company Neura Robotics has created a bipedal robot, 4NE1, weighing 80 kg, which, in the event of a joint failure, does not fall flat but collapses "like a collapsing building," minimizing damage.
Wheels instead of legs: a radical solution
Some developers have gone even further, eliminating the source of risk altogether. Dexmate creates robots on wheeled platforms with long manipulators for working on warehouse shelves. Co-founder Yuzhe Qin explained that the battery and electronics are placed in the platform, ensuring a low center of gravity and eliminating the possibility of falling.
Cobot founder Brad Porter suggests assessing the threat without exaggeration. His wheeled robots with manipulators, pushing carts in hospitals or sorting parts in factories, move at walking speed and do not have super-strong grips. "We don't need to put a lot of energy into the actions. We're not trying to crush watermelons," he noted.
Cryptalist Analysis: While the industry awaits unified safety standards by 2028, we are witnessing a classic "Wild West" in development. Investments in multi-layered protection, from Nvidia's hardware solutions to architectural compromises like wheeled platforms, are not just an ethical issue but a direct factor determining the pace of commercialization. Companies that solve the safety problem first will gain a huge competitive advantage in the B2B market.