Humanoid robotics is the subfield of robotics concerned with the design, construction, and control of robots that exhibit a human-like morphology, typically including a bipedal lower body and articulated upper limbs. Such robots are engineered to operate in environments built for humans, using the same tools, furniture, and physical interfaces. Key technical challenges include stable bipedal locomotion across uneven terrain, dexterous manipulation of varied objects, real-time perception and planning under uncertainty, and safe physical interaction with humans. Advances in machine learning and actuator technology have accelerated commercial deployments in manufacturing, logistics, and care.
Content
- The appeal of humanoid form factors rests on the observation that the human world is built to human proportions: doors, staircases, vehicles, and tools are all designed for bipedal agents with two arms and dexterous hands. A robot that shares this morphology inherits access to an enormous range of existing infrastructure without modification. This insight has driven investment from companies including Boston Dynamics, Agility Robotics, Figure, and Tesla into commercially deployable humanoid platforms.
- Bipedal locomotion remains one of the most challenging control problems in robotics. Unlike wheeled platforms, bipedal robots must continuously manage their centre of mass over a small support polygon, contending with ground contact events, compliance in the terrain, and perturbations. Model predictive control, reinforcement learning trained in simulation, and whole-body controllers that treat the robot as a unified dynamical system have produced robots capable of running, jumping, and recovering from shoves. Sim-to-real transfer remains a bottleneck, as simulation inaccuracies in contact physics degrade policy performance on real hardware.
- Dexterous manipulation is the second major frontier. Human hands contain approximately 27 degrees of freedom and are served by rich tactile and proprioceptive sensing. Current robot hands fall far short of this dexterity, limiting the range of objects and tasks that can be handled reliably. Large-scale imitation learning from teleoperated demonstrations, combined with diffusion policy methods, has produced notable advances in tasks such as folding clothes, assembling electronics, and preparing meals. Standardisation efforts around skill libraries and manipulation benchmarks are helping to accelerate progress across the research community.