3D Collision Cone-based Reactive Avoidance of Multiple Dynamic Obstacles Applied on a Quadrotor Platform

The ability for UAVs to achieve reactive collision avoidance in environments comprising multiple moving objects is an important enabler for UAVs to perform autonomous missions. The relative shapes of the objects play an important role in the determination of appropriate maneuvers for collision avoidance. The literature largely models the shapes of the objects as spheres. However, this can make the avoidance maneuvers very conservative, and this is particularly true when the objects are of elongated shape and/or are non-convex. In this paper, we model the shapes of the objects using suitable combinations of ellipsoids and one-sheeted or two-sheeted hyperboloids, and use a collision cone-based approach to compute collision avoidance maneuvers. The algorithm to construct 3D collision cones for such object shapes is described, followed by computation of an avoidance acceleration law. The algorithm is validated on a PX4 based UAV in ROS-Gazebo, and this is followed by experiments on an actual quadrotor system wherein all the computations for the 3D collision cone and avoidance acceleration laws are performed on an onboard computer in real-time.