Map Optimization of Path Planning in Q-Learning

Abstract

Path planning is significant in the field of artificial intelligence and robotics. This paper proposes a unique map optimization of path planning relying on Q-learning to overcome the shortcomings of classic Q-learning, such as delayed convergence or low efficiency. First, improvements were made to the setup environment, turning a simple environment into a more complex one. Secondly, rewards were set to ensure that each step is optimal exploration. The optimal path is the globally optimal path by setting up, down, left, and right directions simultaneously. Finally, MATLAB simulation was used for verification. As compared to the original training environment, the improved map enhances learning efficiency in a more complicated environment, increases the algorithm's convergence rate, and enables the robot to swiftly discover the collection-free path and finish the job in a complex environment. The rationality of the improvement is verified, which provides important data and a theoretical basis for the subsequent research on Q-learning.