Trajectory-Predicting Network: A Deep Learning Method Enhancing Feasibility of Path Planning

Feasibility of planned path represents the viability and accuracy for unmanned aerial vehicles (UAVs) to track. It is a bridge between UAV path planning and tracking, and also an important guarantee for the safe and efficient completion of the mission of UAV. At present, most researchers only take account of the simplified model and UAV state constraints when considering the path planning, but do not further consider the dynamics constraints of the UAV itself and the performance of the designed path tracking controller, which results in a certain distance error between the planned path and the actual trajectory. When UAVs perform missions in complex and dense areas, the actual trajectory generated by the planned path command even collides with obstacles, which is difficult to guarantee mission safety. This paper proposes a trajectory-predicting network (TPN) based on the deep learning, which characterize the complex nonlinear relation between the planning command and the actual trajectory considering nonlinear model and controller of the closed-loop system. In time of path planning, the predicted trajectory obtained by TPN corresponding to the planning command is evaluated to find the optimal planning command. The optimal trajectory is regarded as a planning path, and its corresponding command as path tracking control input of the control system. Simulation results verify the viability and effectiveness of the proposed method.