Humanoid Robots Show Narrow Skills, Not Multitasking

AFP-JIJI reporting in The Japan Times from the Robotics Summit in Boston in late May documents that humanoid robots can perform headline-grabbing tasks - mixing cocktails, running marathons and folding laundry - while remaining limited in general-purpose multitasking. The piece names prototypes and products including Elon Musk's Optimus, Figure AI's Figure 03, China's AgiBot and Matrix Robotics, and 1X's Neo, which AFP-JIJI reports was steered by a person off to the side. The article includes a direct quote from Chris Matthieu of RealSense: "Most of the humanoids you see are being teleoperated, or they've got very specific paths and chores that they do." William Okazaki of Renesas is quoted saying "I think AI has extremely accelerated that growth." The report also describes advances from a class of systems called VLA models and mentions the idea of a "world model" trained on large image and video corpora (AFP-JIJI in The Japan Times).

Robotics practitioners should read the report as a status snapshot: perceptual and manipulation subsystems are improving, but end-to-end autonomy remains fragile. Companies and labs often combine teleoperation, scripted trajectories and learned components to achieve specific demonstrations. Improving hands, tactile sensing and closing the perception-to-action loop are recurring engineering bottlenecks in public demonstrations.

Observers tracking embodied AI will recognize this as consistent with recent demonstrations that swap generality for reliability. Integration of VLA-style perception with motion controllers is a promising direction, but it amplifies systems-integration and data requirements across vision, language grounding and control.

Indicators that would matter to practitioners include more demonstrations of manipulation under novel perturbations, peer-reviewed evaluations of VLA approaches on robotic tasks, and real-world deployments that operate without persistent teleoperation.