The hierarchical controller based on reinforcement learning algorithm for a multimodule reconfigurable mobile walking robot

Abstract

 Walking robots are able to traverse a variety of obstacles and move around complex landforms. However, the implementation of all the features of the walking configuration is only possible by using a sufficiently complex controller. The article describes a reinforcement learning method applied to solving the problem of traversing obstacles by a reconfigurable walking robot. The algorithm is based on a two-level hierarchical decomposition of the problem, where the high-level controller makes decisions on the trajectory, and the low-level controller moves the limbs into the desired positions. This approach allows the robot to successfully overcome obstacles and move around in unknown environments.