Reactive collision avoidance for small unmanned aerial robots using probabilistic ellipsoidal geometry and bi-directional collision cone

With regard to reactive collision avoidance systems, accurate estimation of the states of dynamic obstacles remains challenging. This paper proposes a reactive collision avoidance system that enables safe autonomous flight of multicopters. The collision avoidance algorithm is designed for a quadcopter platform operating in a three-dimensional dynamic environment with multiple moving obstacles. We incorporate a probabilistic framework within the collision-cone approach to consider prediction uncertainties. The collision probabilities, computed based on the uncertainties in the predicted closest points of approach of dynamic obstacles, are incorporated into the computation of the ellipsoidal bounding box. The size and shape of this bounding box are adaptively adjusted based on the uncertainty, enabling conservative guidance when required and minimizing unnecessary avoidance maneuvers when the uncertainty is low. Further, a bi-directional collision cone is formulated to ensure safe margins even after avoidance, enabling the quadcopter to remain clear of the bounding box. Modified collision cones are constructed in an affine space to reduce computational complexity. Aiming point candidates are determined from the intersections of the collision cones formed by the current position of the quadcopter and its goal point. The final aiming point is selected from these candidates to execute the avoidance maneuver. The collision avoidance algorithm is validated against single and multiple dynamic obstacles via numerical simulations. The simulations confirmed that the algorithm resolved scenarios wherein conventional methods violated the miss distance criterion. The findings highlight that the algorithm affords a robust balance between safety and path efficiency in uncertain environments.