Roboticist Says Safety Must Come First for Humanoid Robots

Videos of humanoid robots have become easy to find online. They show two-legged machines playing soccer, dancing, jumping several feet into the air, and even competing in kickboxing matches. The displays are dazzling, and they have helped humanoid robots capture public attention. Yet for roboticist Aaron Ames, the spectacle raises a troubling question: what happens when these machines leave the lab and enter daily life?

Ames is the Booth-Kresa Leadership Chair and director of Caltech's Center for Autonomous Systems and Technologies (CAST). He is also a leader in the field of robotic safety. His research focuses on controlling complex machines safely. In his view, the most important missing ingredient in humanoid robotics is not agility, intelligence, or appearance. It is safety.

Ames has deep concerns about the current trajectory of humanoid development. He says safety must be the top concern before humanoids can play a larger role in daily life and society. Without a strong foundation in safety, impressive demonstrations may create expectations that the technology cannot yet meet.

The distance between a viral video and a reliable product is considerable. A robot that can jump, kick, or dance in a controlled setting may still fail to predict human behavior, recover from a fall, or avoid a collision. Even a small error can cause injury. Such failures become far more serious when a machine is heavy, powerful, and operating close to people. Humanoids are intended to work in human spaces, which makes physical risk unavoidable unless it is carefully managed.

Safety, Ames argues, cannot be bolted on after a robot is built. It must be a core design requirement from the beginning. That means rigorous testing, verifiable control systems, clear industry standards, and methods that can demonstrate safe behavior across many situations. Caltech's CAST is among the research groups tackling these challenges, combining control theory, robotics, and real-world experiments.

The concern is not limited to a single robot or company. As humanoid robots move from laboratories toward homes, workplaces, hospitals, and public spaces, the safety question will only grow. Regulators, developers, and the public will need ways to assess whether a machine is truly ready to operate around people. Trust, once lost, can be difficult to rebuild. That trust will depend on evidence, not enthusiasm.

Ames's message is direct: humanoid robots may be impressive, but safety must come first. If the field treats safety as an afterthought, broader adoption will remain out of reach. If it treats safety as a foundation, humanoids may eventually deliver on their promise. For Ames, the stakes are clear. The choice, he suggests, will shape how quickly and how widely these machines enter society.