Autonomous obstacle avoidance of underground coal mine transport robots based on intrinsic motivation reinforcement learning algorithm

Existing robot obstacle avoidance methods mostly rely on preset rules or external reward signals, making it difficult to adapt to the complex and variable underground environment in coal mines. To achieve autonomous and efficient obstacle avoidance for underground coal mine transport robots, an autonomous obstacle avoidance method for underground coal mine transport robot based on Intrinsic Motivation Reinforcement Learning (IM-RL) algorithm was proposed. The underground coal mine transport robot perceived external environmental information through visual sensors, calculated internal reward values for identifying external environmental attributes using a curiosity-driven intrinsic motivation orientation function, and computed external reward values for its action attributes using an external motivation reward function. By combining the reward weights of the intrinsic motivation orientation function and the external motivation reward function, it calculated a comprehensive reward value based on the robot's state before and after performing an action, forming the reward mechanism of the reinforcement learning algorithm. The robot's state was trained through a deep belief network, which encouraged the transport robot to actively explore unknown environments. Meanwhile, it used its own memory mechanism to store knowledge and experience, achieving autonomous obstacle avoidance through continuous learning and training. Autonomous obstacle avoidance experiments for the transport robot were conducted in static environments, dynamic environments, and actual underground coal mine environments. The results showed that robots using the IM-RL algorithm achieved the short obstacle avoidance paths and search times, demonstrating strong generalization and robustness.