Abstract: Flying vehicles are a strategic emerging direction leading the future technological development of the automotive field. As one of the mainstream configurations of flying vehicles, split-type flying vehicles are composed of three parts: an autonomous driving chassis, an intelligent cockpit, and a vertical takeoff and landing aircraft. To complete the autonomous docking of the two parts, the autonomous driving chassis needs to track accurately along the planned path to the right below the aircraft. The current motion planning methods lack consideration for many uncertain factors such as sensors, controllers, and actuators, resulting in the chassis trajectory deviating from the planned trajectory during path tracking, making it difficult to accurately travel to the predetermined position and complete docking. To address this issue, an autonomous driving chassis trajectory planning method based on long short-term memory(LSTM) vehicle models is proposed. Using LSTM network to characterize the kinematic characteristics of the autonomous driving chassis of a split-type flying vehicle, a vehicle kinematic model is established based on this. Based on this model, a rolling horizon optimization problem under the model predictive control architecture is constructed. Furthermore, an optimization method based on weighted mean of vectors is used to solve the nonlinear optimization problem and obtain a driving trajectory that conforms to the kinematic characteristics of the chassis. Based on the split-type flying vehicle developed by the team, the proposed planning method is experimentally validated. In the turning scenario, the average longitudinal position deviation, maximum longitudinal position deviation, and longitudinal position docking deviation of the chassis using the proposed method are reduced by 78.89%, 79.64%, and 86.67% compared to traditional MPC methods, respectively.

Key words: split-type flying vehicle, autonomous docking, motion planning, model predictive control