High-speed Safe Trajectory Planning in Confidence-rich Maps

Despite significant progress in active perception methods, one of the greatest challenges toward full flight autonomy of UAVs remains the high-speed flight in unknown and unstructured environments. Leveraging the richer representation of occupancy in confidence-aware maps [1], we address the problem of planning high-speed motions in unknown environments. While considering the future map uncertainty, the proposed simultaneous mapping and planning framework (SMAP) actively reduces the number of surprises, i.e. when the system observes an element of the environment that it did not consider before, leading to fewer trajectory re-planning steps and faster flight with better predictability and higher safety. The key contributions of our approach are: • A thorough analysis of a probabilistic safety measure for trajectories. • Incorporating the confidence-rich map representation into a planning framework that considers future map predictions. • A novel cost function that utilizes the estimate of the covariance of the map, and enables faster and safer plans.