Training UGV For Optimal Path Planning Using Reinforcement Learning

Abstract

In the current era of artificial intelligence technology, unmanned ground vehicles (UGV) are increasingly being widely used in various fields due to the application of advanced technologies such as computer vision and reinforcement learning (RL). These technologies contribute to enhancing the level of automation, environmental awareness, and real-time flexible decision-making capabilities of UGVs. In this paper, we will apply RL with a decaying exploration rate strategy to train UGVs for path planning, obstacle avoidance, and optimal route discovery. Specifically, in the early stages of training, the exploration rate is set to a maximum value so that the agent can explore and collect more information of the state space; as the training time progresses, the exploration rate is gradually reduced, corresponding to a greater focus on the exploitation to find the optimal. Simulation results show that our proposal achieves faster convergence than RL in benchmark, demonstrating the effectiveness of this algorithm for real-world UGV applications.