Reinforcement Learning-Based Intelligent Path Planning for Optimal Navigation in Dynamic Environments

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Path selection and planning are critical for autonomous mobile robots (AMRs) to navigate efficiently and avoid obstacles.
• Traditional methods often use analytical search for shortest paths, but reinforcement learning (RL) offers enhanced performance through action sequence optimization.
• Q-learning, a common RL algorithm, struggles with environment generalization in dynamic systems due to its reliance on cumulative rewards.

Purpose of the Study:

• To optimize reward functions for efficient navigation and obstacle avoidance in RL-based path planning for AMRs.
• To enhance the generalization capabilities of RL algorithms in dynamic environments.
• To evaluate the impact of optimized reward mechanisms on path planning efficiency and learning performance.

Main Methods:

• The study proposes an optimized reward function for RL-based path planning, considering total steps, counted steps, and discount rates in dynamic environments.
• Implemented and analyzed state reward values across different environments using the optimized reward mechanism.
• Evaluated the effect on Q-Learning and Deep Q-Learning algorithms, comparing state-action pair-based performance.

Main Results:

• The optimized reward function significantly decreased the number of iterations and episodes required for learning.
• Achieved a 30% to 70% reduction in overall trajectory distance compared to traditional methods.
• Demonstrated improved path optimization, learning rate, episode completion, and decision accuracy.

Conclusions:

• Optimized reward functions enhance the effectiveness of RL for AMR path planning in dynamic environments.
• The proposed method shows significant improvements in navigation efficiency and obstacle avoidance.
• Combining multiple agents and advanced techniques like federated and transfer learning can further improve convergence and performance on larger maps.