Quadrotor trajectory planning for visibility-aware target following.

In this work, we consider the challenges in the context of the quadrotor trajectory generation for the target-following tasks in cluttered environments where the task may be failed due to the occlusion of the mobile target by structures, i.e. pillars or wall corners. To address this problem, a model predictive control (MPC) based trajectory generation methodology of the quadrotor is proposed to autonomously follow a mobile target with considering flight safety, smoothness, and visibility in cluttered environments. The motion of the quadrotor is formulated as the boundary state constrained primitives (BSCPs), which are constructed offline with the dynamic programming method, and approximated by a pre-trained neural network (NN). Combining with the NN, the proposed method can efficiently generate the following trajectory that explicitly guarantees smoothness and kinodynamic feasibility. The numerical simulation and actual experimental results show that the proposed technique is highly effective. The demonstration video is available at: https://www.bilibili.com/video/BV1iq4y1S7he/.