Abstract:

To improve the path planning capability of mobile robots in a static environment and solve the problem of slow convergence of the traditional Q-learning algorithm in path planning, this paper proposes a multi-step information-aided Q-learning improvement algorithm. Using the multi-step information of greedy action in ε-greedy strategy and length of the historical optimal path to update the eligibility traces, which makes the effective eligibility traces work continuously in the iteration of the algorithm and solves the loop traps that may fall into with the preserved multi-step information; using the local multi-flower pollination algorithm to initialize the Q-value table to improve the robot's pre-search efficiency; based on the purpose of different exploration stages of the robot, the action selection strategy is designed by combining the standard deviation of the iterative path length with the number of times the robot successfully reaches the target point to enhance the algorithm's ability to balance the exploration and exploitation of environmental information. The experimental results prove that the algorithm proposed in this paper has a fast convergence speed, which verifies the feasibility and effectiveness of the algorithm.

Key words: path planning, Q-learning, convergence speed, action selection strategy, grid map