PANTHER: Perception-Aware Trajectory Planner in Dynamic Environments

This paper presents PANTHER, a real-time perception-aware (PA) trajectory planner for multirotor-UAVs (Unmanned Aerial Vehicles) in dynamic environments. PANTHER plans trajectories that avoid obstacles while also keeping them the sensor field of view (FOV) and minimizing blur to aid object tracking. The rotation translation UAV are jointly optimized, which allows fully exploit differential flatness multirotors maximize PA objective. Real-time performance is achieved by implicitly imposing underactuated dynamics through Hopf fibration. able keep inside FOV 7.9 1.5 times more than non-PA approaches decouple yaw, respectively. projected velocity (and hence blur) reduced 18% 34%, leads average success rates three larger state-of-the-art multi-obstacle avoidance scenarios. MINVO basis used impose low-conservative collision constraints position space. Finally, extensive hardware experiments unknown environments with all computation running onboard presented, velocities up 5.8 m/s, relative (with respect obstacles) 6.3 m/s. only sensors an IMU, forward-facing depth camera, downward-facing monocular camera.