Humanoid robots push for general purpose work

Humanoids can walk, but they struggle to work safely around humans.

A panel at the Robotics Summit and Expo in Boston last month laid out the engineering plain truth behind the hype: robotic arms are well within reach for manufacturing, but building a reliable two legged platform that can walk and manipulate objects in a busy environment is still a hard problem. The session brought together a who’s who of the humanoid ecosystem, including Al Makke from Schaeffler, Mike Nielsen of RealSense, Aaron Prather from ASTM International, Alberto Rodriguez of Boston Dynamics, and Pras Velagapudi from Agility, with Mike Oitzman moderating. The event drew about 3,900 attendees to the Boston Convention and Exhibition Center to see everything from component makers to tennis ball shooting robots.

Testing shows that the hardest part is not the leg swing or the arm reach in isolation, but the seamless integration of locomotion, perception, and manipulation in spaces crowded with people, forklifts, and other machinery. The panel stressed that while strides in robotic arms have proven effective for fixed tasks, translating that precision to a mobile, two legged platform creates a new set of failure modes. In particular, designers must contend with stability on varied floors, fast-changing scenes, and the need to avoid unintended contact with nearby humans and objects. The realities of perception and control in dynamic environments were repeatedly highlighted as the main roadblocks to turning demonstrations into dependable operation.

Boston Dynamics figures prominently in the discussion as the Atlas program’s stated goal, a general purpose machine tailored to physical labor. Alberto Rodriguez emphasized that customers are pushing for robots that can adapt to a broad range of tasks rather than excel at a single chore. That push toward generality brings tradeoffs in control complexity, energy use, and reliability, but it also aligns with a broader industrial demand for adaptable automation that can reduce the need for bespoke solutions for every function.

The session also highlighted the ecosystem around humanoids. RealSense contributes sensing capabilities that are essential for perception in cluttered workspaces, while Schaeffler and other components suppliers are refining joints, actuators, and control electronics to shrink size, weight, and energy consumption. ASTM International was represented to signal that standardized testing and shared benchmarks are entering the conversation, a trend many engineers see as vital for comparing progress across vendors and for accelerating real world adoption.

Demos at the expo underscored how far the field has come and how far it still has to go. Attendees could see everything from vendor hardware to experiments aimed at more playful tasks like ball games, but the underlying takeaway remained pragmatic: a humanoid that can walk is not yet a guaranteed worker. The panel underscored a shift from spectacular demos to a more grounded, customer-driven development arc, with engineers collaborating to prove stability, safety, and repeatability in real workplaces.

Industry watchers will want to monitor two dynamic tensions as this work continues. First, the balance between general purpose capability and task-specific reliability, the more flexible the robot, the more demanding the control problem becomes. Second, the maturation of industry standards and shared testing protocols, which could unlock faster qualification of humanoids by reducing the risk and cost of fielding new designs. If the Summit’s takeaways hold, the path forward is less about wow moments in demos and more about measurable, dependable performance in real environments.