The Long Road to Humanoid Robots: From Walking Machines to AI Workers

Waseda University researchers built WABOT-1 to combine several abilities in one machine: walking on two legs, seeing with a rudimentary vision system, gripping simple objects, and communicating in basic spoken Japanese. It wasn't practical or fast. But it reframed the humanoid robot as a system that had to coordinate multiple senses and actions together, not a single mechanical trick like an isolated walking leg or a robotic arm.

Honda packed batteries, computers, and actuators inside P2's own body, letting it walk and climb stairs without a tethered power or control cable. It wasn't autonomous in how it made decisions by today's standard, but cutting the cord proved a humanoid could carry its own power and processing rather than depending on an external umbilical.

Honda introduced a smaller, more refined ASIMO, with better balance and motion prediction while walking. It became the public face of humanoid robots for over a decade, touring stages worldwide. But every smooth demo depended on a carefully prepared environment and a scripted task, not on the robot adapting to whatever it happened to encounter.

Boston Dynamics entered Atlas in DARPA's Robotics Challenge, which tested tasks like opening doors, using tools, and moving through rubble, the kind of settings too dangerous to send people into. The competition showed real progress. It also produced a lot of falls, and made clear that tasks trivial for a human, like turning a valve or climbing out of a car, could be extremely hard for a robot.

Tesla announced it wanted to carry its computer-vision and manufacturing expertise over from cars into a general-purpose humanoid robot, built to take on repetitive or dangerous work. The announcement widened commercial interest in the whole category. What it showed on stage, though, was a person in a robot costume, not a working prototype, a roadmap rather than evidence.

Figure launched with a plan to build general-purpose humanoid robots for jobs facing labor shortages or safety risks. Fast funding rounds and early industrial partnerships signaled that investors and manufacturers saw a plausible commercial opening. But a successful short demo is a long way from proving a robot is economically worth running every single day.

Boston Dynamics retired the hydraulic Atlas that had defined a decade of disaster-response research, and introduced an all-electric successor aimed at commercial and industrial work. Electric actuators give the robot more freedom in joint design and could simplify maintenance. But the promotional footage of it moving didn't prove it can manipulate objects reliably or work safely around people at any real scale, that's still being tested.

Figure delivered its second-generation robot to BMW's Spartanburg plant, then spent months testing before running it on an active welding line every working day for the next ten months. Figure 02 picked up sheet-metal parts and placed them within a 5-millimeter tolerance in under two seconds, a task demanding enough that BMW measured it against strict cycle-time and accuracy targets. The deployment ended not because the robot failed, but because its successor, Figure 03, was ready to take over.

UC San Diego engineers and surgeons tested two humanoid robots, nicknamed Surgie, inside an operating room built for people rather than a specialized surgical rig. In one procedure, a robot worked alongside a human surgeon to remove a gallbladder. In a second, two robots operated together with no human in the loop. The robots needed several mid-surgery recalibrations, so both procedures took longer than they would have on dedicated surgical systems, but researchers say the precision matched existing robotic surgery platforms at a fraction of the cost and footprint.