The humanoid robot has been a fixture of science fiction for decades and a recurring punchline in robotics for almost as long. Expensive, fragile, slow, and capable only in controlled laboratory conditions. That characterization is becoming outdated. Not because the fundamental engineering challenges have been solved — they have not — but because the combination of improvements in AI control systems, actuator technology, and training pipelines has moved humanoid robots from demos to limited production deployment.
Figure AI's partnership with BMW represents one of the clearest real-world data points. Figure 01 robots were deployed at a BMW manufacturing facility in Spartanburg, South Carolina, handling parts transfer and assembly tasks alongside human workers. The deployment was limited in scope and closely supervised, but it represented a genuine production environment rather than a staged demonstration. Figure 02, the company's second-generation platform, showed significant improvements in manipulation dexterity and task completion speed in company demonstrations.
Tesla's Optimus program has followed a similar trajectory. Elon Musk has described Optimus working in Tesla's own factories, sorting battery cells and performing simple manipulation tasks. The claim that Optimus would be available for sale as a general-purpose robot at a consumer price point has been consistently deferred, but the internal factory deployment appears to be real progress. Tesla's advantage is its existing manufacturing scale, which provides both a training environment and a potential deployment channel that no pure-play robotics company has access to.
Boston Dynamics' Atlas, now on its electric fourth-generation platform, brings decades of bipedal locomotion expertise. The electric Atlas moves with a fluidity that hydraulic systems could not achieve, and its integration with AI perception systems has advanced significantly. Boston Dynamics' commercial focus, however, has remained on Spot — its four-legged inspection robot — which has found substantial adoption in industrial inspection, hazardous environment monitoring, and construction site surveying.
The honest assessment of where humanoid robotics stands: the locomotion problem is largely solved for structured environments. Walking, climbing stairs, standing up from falls — these capabilities are real. The dexterous manipulation problem — picking up arbitrary objects, handling fragile items, operating tools not designed for robot hands — remains genuinely difficult and is the primary bottleneck for general manufacturing deployment.
The economics are also still challenging. A humanoid robot in the current generation costs significantly more than a human worker over a multi-year horizon when you include the cost of the robot, its maintenance, and the engineering supervision required. The investment case is most compelling for tasks that are dangerous for humans, where human labor is genuinely scarce, or where 24-hour operation provides a return that human shift scheduling cannot match.