Research on the Effect of Vehicle Obstacle Avoidance Problem Based on Q-learning

Machine learning is one of the key topics for researchers. As an important branch of machine learning, reinforcement learning focuses on how agents learn strategies through trial and error in dynamic environments. The introduction of Q-learning was a major contribution to value-function-based methods in reinforcement learning. This passage uses a system of controlling variables for simulation and analysis based on Matlab Robot Playground to conduct research on the influence of parameters in Q-learning under different obstacle conditions and try to find the optimal choice of parameters in different situations. The two most important variables in this study are the learning rate α and the discount factor γ. When the alpha value is higher, the robot's path is simpler, and the learning takes a shorter time. The effect of the size of the gamma value on path planning is not regular. However, it can be found that the bigger the value is, the path choice of the agent will favor experience in the past. This study starts with the surface functions of the Q-learning algorithm and discusses the impact of major variables on the algorithm results, providing a unique approach for in-depth research on Q-learning algorithms.