Path Planning and Obstacle Avoidance of a Climbing Robot Subsystem using Q-learning

This article discusses the intelligent path planning of the serial subsystem of the scaffolding climbing robot that climbs the scaffolding structure amidst the presence of some scaffold obstacles. Q-learning algorithm is currently used in computational learning. Therefore, reinforcement learning was chosen to intelligently design the robot's path and obstacle avoidance given its specific mechanical and environmental conditions. It is noteworthy that to achieve this goal, speed is also adopted as one of the state variables in path planning under intelligent training. To ensure successful learning, an effective and optimized reward function and strategy have been chosen and implemented. For the serial subsystem in this robot, Q-learning and Double Q-learning methods were used to obtain intelligent path planning results by simulation. Comparison and analysis have been done for the computation time of both approaches and the speed of convergence of their respective algorithms.