Footstep Planning Using Learning-Based MPC in Limit-Cycle-Based Walking

Abstract

In this study, Passive Dynamics Autonomous Control (PDAC) is handled within the framework of model-based reinforcement learning to achieve footpstep planning. By introducing a double support phase into PDAC, the amount of conserved quantities are adjusted to stabilize gait control. Afterwards, neural networks are used to learn the transition of the conserved quantities depending on the footstep and their target values as a dynamics model. Using the dynamics model obtained from the learning, it is possible to execute the footstep planning considering the states of multiple steps ahead. In the experiment, when a desired goal position was given, the behavior of the robot approaching the goal position was confirmed within the range in which walking can be stabilized.