Vision-Based Reactive Planning for Aggressive Target Tracking while Avoiding Collisions and Occlusions

International audience—In this paper we investigate the online generation of optimal trajectories for target tracking with a quadrotor while satisfying a set of image-based and actuation constraints. We consider a quadrotor equipped with a camera (either down or front-looking) with limited field of view. The aim is to follow in a smooth but reactive way a moving target while avoiding obstacles in the environment and occlusions in the image space. We propose vision-based approaches based on multi-objective optimization, especially with the occlusion constraint formulation. We design an online replanning strategy inspired from Model Predictive Control (MPC) that successively solves a non-linear optimization problem. The problem is formulated as a Nonlinear Program (NLP) using differential flatness and finite parametrization with B-Splines. This allows a resolution by Sequential Quadratic Programming (SQP) at a rate of 30Hz. The robustness and reactivity of the replanning algorithm are demonstrated through realistic simulation results. Experiments validating the performance with a real quadrotor are also presented