Multi-Rotor UAV Trajectory Planning in Warehouse Picking: An Imitation Learning Approach with Human Directional Correction

Existing path planning algorithms struggle to generate paths that better align with human preferences and requirements when applied to path planning tasks in complex warehouse environments. Imitation learning methods for human directional corrections in drone-based warehouse picking tasks may face significant challenges when applied to the requirement of maintaining a reasonable flight height. Due to the lack of relevant environmental constraints, the number of iterations required may significantly increase. Additionally, when precision in reaching target points for picking is required, existing imitation learning methods may exhibit an increasing trend in terminal position errors as the number of corrections increases. This study employs a human directional correction-based imitation learning method to implement trajectory planning in dynamic warehouse environments, ensuring that the generated trajectories align more closely with the user's expectations and intuition. The method incorporates a cost function to maintain reasonable flight height and introduces an error coefficient that increases with the number of corrections. Through the construction of a three-dimensional warehouse simulation platform and subsequent simulation verification, the results demonstrate that an average of only six iterations is required to meet basic picking requirements. This paper offers support for the enhancement of warehouse logistics automation.