A Study of PPO Algorithms Combining Curiosity and Imitation Learning in Maze Environments

This study implements a maze navigation robot on the Unity platform, aiming to solve the problems of sparse rewards and slow model convergence, which are common in maze environments. To address the problem of sparse rewards, the study adds an intrinsic curiosity mechanism to the baseline algorithm PPO, which compensates for the lack of external rewards by incentivizing the agent to explore unknown regions. Meanwhile, in order to accelerate the convergence process of the model and improve its stability, the study further combines deep reinforcement learning techniques with imitation learning to optimize the model. Experimental results show that the proposed method not only significantly improves the learning efficiency of the model in the maze environment, but also exhibits high robustness and adaptability.