Path Planning and Energy Optimization in Optimal Control of Autonomous Wheel Loaders Using Reinforcement Learning

Abstract

This paper proposes a novel solution based on reinforcement learning for optimal control of an autonomous Wheel Loader (WL). The solution considers the movement of a WL in a Short Loading Cycle (SLC) as a switched system with controlled subsystems such that the sequence of active modes is fixed. Therefore, the optimal control system solves two different levels of optimization. In the upper level, optimal switching times are sought. In the lower level, the control inputs to navigate the wheel loader and performing path planning are sought. For solving the problem, Approximate Dynamic Programming (ADP), which is the application of reinforcement learning to find near-optimal control solution, is used. Simulation results are provided to show the effectiveness of the solution. At last, challenges of using the proposed method and future works are summarized in Conclusion.