Online quadrotor trajectory generation and autonomous navigation on point clouds

Fei Gao, Shaojie Shen

Abstract: In this paper, we present a framework for online generation of safe trajectories directly on point clouds for autonomous quadrotor flight. Considering a quadrotor operateing in unknown environments, we use a 3-D laser range finder for state estimation and simultaneously build a point cloud map of the environment. Based on the incrementally built point cloud map, we utilize the property of the fast nearest neighbor search in KD-tree and adopt the sampling-based path finding method to generate a flight corridor with safety guarantee in 3-D space. A trajectory generation method formulated in quadratically constrained quadratic programming (QCQP) is then used to generate trajectories that constrained entirely within the corridor. Our method runs onboard within 100 milliseconds, making it suitable for online re-planning. We integrate the proposed planning method with laser-based state estimation and mapping modules, and demonstrate the autonomous quadrotor flight in unknown indoor and outdoor environments.