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. We propose Predictive Style Matching, in which an offline predictor maps the robot's lower-body state history and velocity commands to interpretable upper-body joint and gait targets that shape the rewards during training. Because the targets are state-conditioned rather than time-indexed and the predictor is used only at training time, the deployed controller inherits the proprioceptive interface and inference cost of a task-only RL baseline. On the Unitree G1, in both simulation and hardware, PSM reduces upper-body style error by roughly an order of magnitude over task-only RL while preserving its fall-recovery rate, whereas the motion-imitation baseline attains the lowest style error but fails to recover from disturbances about five times as often.

VIMPPI: Enhancing Model Predictive Path Integral Control with Variational Integration for Underactuated Systems

This paper presents VIMPPI, a novel control approach for underactuated double pendulum systems developed for the AI Olympics competition. We enhance the Model Predictive Path Integral framework by incorporating variational integration techniques, enabling longer planning horizons without additional computational cost. Operating at 500-700 Hz with control interpolation and disturbance detection mechanisms, VIMPPI substantially outperforms both baseline methods and alternative MPPI implementations