Perception-Constrained Vision-Based Quadrotor Control

One of the most critical issues in vision-based quadrotor control is to keep the target within the camera field of view. This is challenging in aggressive flights since the quadrotor is an under-actuated system that cannot translate without rotation. We present a perception constrained vision-based controller for quadrotors that can ensure target visibility. Our framework leverages nonlinear model predictive control to formulate a constrained visual servo control problem, where the actuation limit and target visibility are explicitly considered. We utilize image error instead of position error to accomplish the target tracking tasks which make our system robust to camera calibration error. Simulation results show that our method can work without the need for positional information, and it successfully completes the target tracking tasks from multiple initial configurations while keeping the target visible. It verifies the effectiveness of the defined perception constraint in aggressive quadrotor maneuvers.