Towards the Realtime Sampling-based Kinodynamic Planning for Quadcopters

This paper presents a sampling-based kinodynamic planning algorithm for quadcopters which have a high dimensional state space amid state constraints. The proposed approach utilizes BIT* as the main planner to perform the ordered search directly in state space, without decoupling the geometric and dynamic constraints. Instead of using a general BVP solver for non-trivial local steering problem, a neural network based approximation is adopted to boost the computational efficiency. The potential of the proposed approach towards a real-time onboard implementation is demonstrated by various simulations and a real flight test.