As companies race to bring human-shaped machines into factories, offices and homes, safety systems built on sensors, software and engineering safeguards are becoming the industry’s most urgent priority.

SAN FRANCISCO — The race to build humanoid robots is moving quickly from spectacle to deployment, but the industry’s biggest challenge may not be making machines that walk, lift or learn. It is making them safe enough to work beside people.

Technology companies and robotics startups say a new generation of electronics, sensors and artificial intelligence will allow humanoid robots to operate in shared human environments, from warehouses and factories to offices and eventually private homes. But recent mishaps and demonstrations have renewed concerns about the risks posed by machines that can be heavy, mobile and increasingly autonomous.

Unlike traditional industrial robots, which often work behind cages and perform tightly scripted movements, humanoid robots are designed to move through the same spaces as humans. That ambition creates a more difficult safety problem. These machines must recognize people, predict motion, avoid collisions, react to unexpected obstacles and stop safely when something goes wrong.

Nvidia has entered the field with safety technology aimed at giving robots faster decision-making ability in human environments. The company says humanoid machines will need to make split-second judgments based on what is happening around them before they can be trusted to interact closely with people.

Other companies are pursuing layered safety systems. Fort Robotics and Nvidia are among firms developing technologies intended to combine sensors, computing and real-time hazard detection so robots can identify dangerous situations before physical contact occurs. The Wall Street Journal reported that international standards specifically addressing humanoid robot safety are expected by 2028, underscoring how quickly the sector is moving ahead of formal regulation.

The urgency is growing because humanoids are no longer confined to research labs. Apptronik, a Texas robotics startup valued at more than $5.5 billion, has opened a nearly 90,000-square-foot training facility in Austin where its Apollo robots practice tasks such as loading boxes and sorting objects. The company is using those exercises to generate data for future factory and service applications.

Industrial deployment is already taking shape. British robotics company Humanoid plans to deploy between 1,000 and 2,000 robots at facilities operated by German supplier Schaeffler by 2032, beginning with test uses in box handling and manufacturing integration in Germany.

Yet the closer robots move toward people, the more complex the safety equation becomes. A robot working in a factory may face predictable routes, trained workers and controlled tasks. A robot in a home or office must deal with children, pets, stairs, clutter, sudden movements and people who may not understand how the machine behaves.

That is why some companies are rethinking the humanoid form itself. Instead of building fully legged machines that can fall or lose balance, some manufacturers are using wheels, softer bodies or designs that reduce impact risk. Weave Robotics, for example, has introduced Isaac 1, a wheeled home robot designed for basic chores such as laundry, choosing stability and a softer appearance over full human imitation.

Others are trying to make human-shaped robots less dangerous by changing how they move. Engineers are focusing on lighter materials, force limits, emergency stop systems, joint monitoring, soft coverings and software that can slow or freeze motion when people enter a robot’s safety zone. Research in human-robot interaction has long emphasized collision avoidance, reducing the force of impacts and emergency response as core safety layers.

The consumer market remains especially uncertain. 1X Technologies has promoted its Neo humanoid robot for home use, but reporting on the company has highlighted how much such systems may still depend on human teleoperators and how difficult full autonomy remains in everyday environments.

That dependence on remote control raises not only safety questions, but privacy and trust concerns. If a robot in a home still requires human assistance behind the scenes, users may need to accept that some tasks are being observed or guided by outside operators. The boundary between helpful machine and monitored household device could become a central issue as domestic robots move toward commercialization.

Supporters argue that the technology can ultimately make work safer by taking over hazardous jobs. Robots could handle heavy lifting, repetitive movements, chemical exposure or dangerous inspections, reducing injuries among human workers. But that promise depends on whether the robots themselves can be made predictable, reliable and fail-safe.

The industry’s central dilemma is therefore clear: humanoid robots are being designed to enter human spaces precisely because those spaces were built for human bodies. But the more human-like and mobile the machines become, the more they must prove they can operate without becoming a new source of risk.

For now, the future of humanoid robotics rests less on dazzling demonstrations than on a quieter engineering challenge: teaching machines not just how to act like humans, but how to behave safely around them.