Deep-reinforcement-learning-based trajectory tracking control for slidable-wheel omnidirectional mobile robot

Abstract

Deep reinforcement learning (DRL) has been widely applied to robotic control, with trajectory tracking control being a particularly popular topic. However, existing research primarily focuses on simple models with low degrees of freedom (DOFs). The slidable-wheel omnidirectional mobile robot (SWOM), previously proposed by the authors, is an omnidirectional mobile robot characterized by passive moving components and 12 DOFs. The complexity of this structure presents significant challenges for controlling SWOM using DRL. In this paper, we propose a control framework that integrates DRL with hierarchical control to achieve DRL-based trajectory tracking for SWOM. The upper-level controller computes the required velocity of the robot body based on the error between the current and target positions, thereby regulating the wheel’s rotational speed. The lower-level controller consists of three DRL-trained sub-controllers, each responsible for steering one of SWOM’s three wheels, ensuring that the passive components of SWOM remain within an acceptable range. The design of the reward function in reinforcement learning, as well as the impact of parameter variations on training outcomes, is discussed. The effectiveness of the proposed controller is validated through numerical simulations conducted under two scenarios: one without considering wheel slippage and the other with wheel slippage taken into account. The proposed controller demonstrates satisfactory performance in both cases.