Abstract
Abstract Abstract-This paper describes on the development of a tendon-driven biped robot that can absorb impact forces when it is walking, running, and so on. Furthermore nonlinear spring tension devices (NST) named mpNST are attached at the end of tendons.This device makes it possible to adjust the mechanical joint stiffness and guarantee the robustness for the impulsive reaction forces from the ground. Next, this paper describes on reinforcement learning of rules to correct the waist trajectory and tendon- tensile forces for stable walking by Improved Penalty Avoiding Rational Policy Making algorithm (PARP) based on Profit Sharing method (PS). Some simulation results are given.
