Real-time optimal collision avoidance for multiple UAVs: An embedded CCP-ADMM approach

초록

This paper presents a real-time trajectory optimization framework for multi-UAV collision avoidance designed for embedded implementation. Optimization-based planners can explicitly handle dynamics and safety constraints, but repeated onboard solves remain computationally demanding. To reduce this burden, we combine the convex-concave procedure (CCP) for non-convex collision avoidance constraints with a projection-oriented ADMM formulation based on reusable linear-system updates and closed-form radial and pairwise half-space projections. The proposed CCP-ADMM planner is benchmarked against ORCA and a MOSEK-based sequential convexification baseline in head-on and intersection scenarios, and further evaluated through K-scaling Monte Carlo tests using runtime, normalized control effort, constraint satisfaction, and success-rate metrics. Processor-in-the-loop simulations on NVIDIA Jetson Orin Nano boards validate embedded execution and replanning, while indoor quadrotor flight experiments demonstrate practical safety-distance maintenance on a physical testbed. The results indicate that the proposed framework provides a practical balance between trajectory-level constraint handling and embedded computational efficiency for multi-UAV collision avoidance.

키워드