World Model-based Deep Reinforcement Learning Considering Timings of Touchdown and Liftoff

Abstract

In this study, we aim to develop a human-like biped robot by integrating the world model-based deep reinforcement learning with the passive dynamical mechanism that consists of the interaction between the body and the environment. Combining deep reinforcement learning with central pattern generator (CPG), which is the nervous system that generates rhythms for human locomotion, learns human-like periodic movements. Specifically, the robot learns a jumping movements from high-speed camera images of itself and CPG phase, taking into account the timings of liftoff and touchdown. In the actual experiment, a learning method called ”DreamerV2” was used to obtain periodic continuous jumping movements.