Collision-Free Trajectory Generation for Multiple UAVs with Sensing Constraints

This paper mainly focuses on the real-time collision avoidance trajectory generation for multiple UAVs with limited sensing capabilities. It is assumed that each UAV has visual constraints (i.e., limited field of view (FOV)). Based on their sensed relative positions, two observation situations are considered: reciprocal observation and unidirectional observation, which can be identified through our proposed observation function. We propose a two-stage strategy for each UAV to plan its real-time trajectory, resulting a very small deviation from the initial trajectory. At the first stage, the quadrotor employs a novel visual vector field function to judge whether a collision possibly occurs, then a modified optimal reciprocal collision avoidance (ORCA) algorithm is proposed such that collision-free zones are figured out. At the second stage, the quadrotor will regenerate a smooth trajectory by solving a QP problem to return to the initial trajectory. Simulation and experiment results show the validness and effectiveness of our proposed strategy.