Achieving Maximum Mobility and Manipulation Using Human like Compliant Behavior and Behavior Libraries
Technical Report,21 Feb 2011,20 Aug 2015
Carnegie Mellon University Pittsburgh United States
Pagination or Media Count:
We implemented balance, walking, and push recovery on our humanoid robot. We developed a new approach to efficient robust policy design. We developed efficient algorithms to calculate first and second order gradients of the cost of a control law with respect to its parameters, to speed up policy optimization. This approach achieves robustness by simultaneously designing one control law for multiple models with potentially different model structures, which represent model uncertainty and unmodeled dynamics. We developed a footstep planning approach that takes into account robot dynamics. We showed that optimal stepping trajectories and trajectory cost for a walking biped robot can be encoded as a simple function of initial state and footstep sequence.