Human-intuitable Collision Avoidance for Autonomous and Semi-Autonomous Aerial Vehicles

This paper addresses the problem of path planning and collision avoidance for multiple aerial vehicles. We develop an algorithm that is scalable, operates in real-time, and is implementable on a UAV's onboard computers. Our method makes use of velocity-based potential field methods for collision avoidance. Potential field methods utilize attractive and repulsive potential to guide the robot towards the goal. We demonstrate the system in a simulation involving 50 vehicles. Next, on physical platforms, we conduct experiments with two and three UAVs capturing the system's ability to avoid other moving vehicles and measuring closest approach. We find that the UAVs not only avoid the collisions but also maintain a minimum distance specified. This method can be used in the future for trajectory planning of multiple aerial vehicles in dense airspaces.