Predictive Style Matching: Natural and Robust Humanoid Locomotion
Reinforcement learning has become the prevailing approach to humanoid locomotion control: policies transfer reliably from simulation to hardware and recover gracefully from disturbances. Motion quality, however, still lags behind: task-only rewards often converge to stiff, asymmetric gaits, while motion imitation methods improve appearance but become more sensitive to external disturbances because reference signals can oppose the transient poses needed to regain balance.
Evidence notes
- Experiments use G1 as the physical evaluation platform.
- The reported results provide an independent evaluation on G1 hardware.
Company context
Develops quadruped and humanoid robots, with strength in dynamic locomotion, vertically integrated hardware, and relatively low-cost commercial deployment. Unitree is one of the clearest examples of a legged robotics company moving from research visibility into real productisation and broader market distribution.