Assembly Measurement Path Planning for Mobile Robots Using an Improved Deep Reinforcement Learning

In addressing the challenges associated with mobile robot path planning during complex product assembly measurements, this study introduces an N-step Priority Double Q Network Deep Reinforcement Learning algorithm (NDDQN). To enhance the algorithm's convergence speed, we employ double Q-learning and an N-step priority strategy during the learning phase. This approach aims to improve the obstacle avoidance capabilities of mobile robots while accelerating their learning efficiency. We conducted three grid-based obstacle avoidance simulation experiments of varying scales to compare and analyze the path planning performance of both the proximal policy optimization algorithm and the Deep Q Network algorithm. To accurately simulate real-world robotic measurement scenarios, two Gazebo environments were utilized to validate the effectiveness of our proposed algorithm. Through a comprehensive analysis of simulation results from all three algorithms, we demonstrate that the NDDQN algorithm exhibits significant effectiveness and stability in path planning. Notably, it substantially reduces iteration counts and enhances convergence speeds. This research provides a theoretical foundation for adaptive path planning in mobile robots engaged in complex product assembly measurements.