A Novel 3-Dimensional Crane Lifting Path Planning Method Based on Reinforcement Learning

Lifting operation planning of mobile cranes has relied on the experience of operators for a long time, leading to inefficiencies and compromised safety. Consequently, an increasing number of algorithms are being applied to crane lifting operation planning. However, operation paths planned by these algorithms are often not adopted by operators in practical applications due to complexity and insecurity. This paper aims to integrate the advantages of experienced operators’ habits into the planning algorithms, thereby enhancing operational safety and reducing complexity while ensuring planning efficiency. This integration results in more practically significant operation paths. We explore the integration of the A* algorithm with Q-learning, aiming to create smooth operational sequences for crane manipulator path planning within the configuration space. We adopt a strategy that minimizes the number of action switches, utilizing improved Q-learning for global lifting operations and the A* algorithm for motion planning. This algorithmic framework addresses the limited adaptability of Q-learning in dynamic environments and has been proven to successfully generate a collision-free, smooth operation sequence. The findings highlight the potential of reinforcement learning in crane operation planning, suggesting future integration with traditional methods to optimize path quality and efficiency.