Modified Reinforcement Learning With Pid Control for Obstacle Avoidance in Autonomous Robots

This paper investigates the obstacle avoidance problem for autonomous robots, a critical challenge in dynamic and uncertain environments. We propose a modified Q-learning algorithm integrated with PID control to achieve dynamic and adaptive navigation for collision avoidance. The modified Q-learning incorporates two novel elements: a riskbased penalty and an exploration penalty, which enhance the learning process. The risk penalty discourages actions that lead to unsafe proximity to obstacles, ensuring safer navigation. The exploration penalty optimizes the balance between exploration and exploitation, enabling the agent to focus on actions with higher confidence while still adapting to new situations when necessary. By combining reinforcement learning for high-level decision-making and PID control for fine-tuned action execution, the proposed algorithm enables robust and efficient obstacle avoidance, enhancing safety and adaptability in real-world scenarios.