Humanoid Loco-Manipulation Policy Learning

RL-based whole-body control for locomotion, reaching, and motion tracking.

This project focuses on scalable policy learning for humanoid robots that must walk, manipulate, and track motion under realistic dynamics.

Key contributions:

  • Built RL pipelines for robust whole-body control across locomotion and object-reaching tasks.
  • Improved policy quality and adaptability using teacher-student distillation from retargeted human demonstrations.
  • Designed adaptive behaviors such as low-object reaching with natural full-body motion.
  • Prioritized sim-to-real robustness to transfer policies from simulation to physical humanoid platforms.

Technical stack:

  • IsaacLab / IsaacGym
  • PyTorch
  • ROS