Decentralized receding horizon control for multiple UAVs

This chapter presents recent work on the design and implementation of on-line trajectory optimization algorithms on our multi-vehicle testbed. This work extends the previous receding horizon control (RHC) for a single vehicle to handle scenarios with multiple vehicles by explicitly including collision avoidance constraints in a distributed planning system. The basic RHC trajectory design problem is encoded as a mixed-integer linear program (MILP), but this optimization is only solved for a detailed trajectory that extends part of the way towards the target waypoint. The rest of the trajectory is represented by an approximate cost-to-go function. This RHC approach enables us to exploit the power of the MILP formulation to encode the collision and obstacle avoidance constraints in a computationally tractable algorithm. However, even the solution times of the RHC scale poorly with the fleet size. To resolve this problem, we developed a technique for embedding the collision avoidance constraints in a distributed formulation of the RHC. In the new approach, vehicles plan their own trajectories using RHC while analyzing the published plans for conflicts. The reaction to a detected conflict is to solve a coupled MILP optimization that explicitly includes a cooperative maneuver. Real-time tests of this overall control system on the hardware testbed show that this approach could be scaled to much larger teams with a very small degradation in the performance.