A residual convolutional neural network based approach for real-time path planning

Path planning for unmanned aerial vehicles (UAVs) has been widely considered in various tasks. Existing path planning algorithms, such as A* and Jump Point Search, have been proposed and achieved good performance in the static mode, that is, assuming the global environmental information is known and planning is conducted offline. However, in practice, only limited environmental information can be obtained by sensors, which requires a real-time path planning ability. This paper proposes a residual convolutional neural network based approach, denoted as Res-Planner, to address the real-time path planning problem of a UAV. Specifically, the approach generates various scenarios and paths by executing the conventional path planning algorithms in static mode, from which it collects state-behaviour demonstrations to train the proper behaviour in real-time path planning. The experimental results show that our approach can provide feasible paths with approximately the global optimal under limited environmental information conditions.