The document discusses control of complex postural tasks for humanoid robots based on learning from demonstration. It covers several topics: imitation learning and skill innovation for single skills; planning and control of sequential skills; postural control using fractional calculus controllers; and high-level control architecture for executing complex tasks. The key contributions are in postural motion planning and control using reward functions; modeling sequential skills as reward transition matrices; and applying fractional control to reduce model disturbances. Future work areas include goal emulation, reward profile selection using machine learning, and implementing fractional control on a real humanoid.
Related topics: