Research on Mobile Robot Path Planning Based on Meta-DQN Algorithm

Currently, artificial intelligence shows vigorous development. Mobile robot path planning, as a fundamental research topic in the field of artificial intelligence, is constantly undergoing iterative upgrades every year. DQN, as a traditional algorithm for deep reinforcement learning, has many limitations: the training process is trial and error by interacting with the environment intermittently, but every time it finishes training, the model needs to start from scratch to train the next task, which results in low-quality samples, long training time, and poor robustness. To solve the above problems, this study adds meta-learning on the basis of the traditional DQN algorithm to form a meta-depth reinforcement learning improvement algorithm, which takes advantage of their strengths and complements their weaknesses, utilizes the ability of meta-learning to adapt to the new task quickly, and fine-tunes the model through meta-knowledge so that the model itself obtains an autonomous learning mechanism, which greatly reduces the time to train the model. The environment was set as a raster map, the starting point, end point, and roadblock were determined, and the simulation was carried out on the MATLAB platform. The results show that the improved algorithm possesses the advantages of reducing the frequency of interaction with the environment, arriving at the goal point smoothly, and taking a shorter time compared with the traditional DQN algorithm.