Model Predictive Control of Quadrotor Swarms via Computationally Tractable Collision Avoidance Constraints

This paper presents a model predictive control (MPC) algorithm with computationally tractable constraints for coordinating a quadrotor swarm relying exclusively on onboard sensing. The proposed formulation includes continuous-time collision checking that enables real-time optimization on inexpensive hardware without external localization systems. Experimental validation with Crazyflie 2.1 drones demonstrates that obstacle avoidance, inter-drone separation, and coordinated target assignment can be achieved despite limited sensing accuracy and communication bandwidth. A quantitative analysis of estimation uncertainty, computation time, and communication load highlights the practical constraints and scalability limits of this centralized MPC implementation. The results show that reliable multi-drone navigation can be accomplished with minimal hardware requirements, making this approach suitable for low-cost deployment scenarios.