Reinforcement Learning of Stable Trajectory for Quasi-Passive Dynamic Walking of an Unstable Biped Robot

Dynamic Control Algorithm for Biped Walking Based on Policy Gradient Fuzzy Reinforcement Learning

“…The proposed Reinforcement learning structure is based on policy gradient Methods (Peters et al [2003], Shibata et al [2007], Tedrake et al [2004]). The policy-gradient method is a stochastic gradient-descent method.…”

“…In this paper, we use a policy-gradient method for learning efficient biped motion. The policy-gradient method is a kind of reinforcement learning method which maximizes the average reward with respect to parameters controlling action rules known as the policy (Shibata et al [2007], Tedrake et al [2004], Peters et al [2003]). In comparison with most standard value function-based reinforcement learning methods, this type of method has particular features suited to robotic applications.…”