Simulation study demonstrates zero-collision multi-UAV navigation across congested dynamic 3D environments, highlighting the value of priority-aware temporal coordination.
Key Points
To develop and evaluate a multi-UAV navigation framework that prevents deadlocks, persistent yielding, and congestion in dynamic environments without relying on static planning assumptions.
Designed T-CARE, combining zero-shot constrained action selection with runtime spatiotemporal reservations, priority aging, stagnation recovery, and bottleneck corridor reuse.
Tested the framework under zero-shot deployment in simulations featuring three-swarm adversarial stress tests and ten-swarm (40 total UAVs) 3D urban and suburban environments.
T-CARE achieved a 100% success rate and a 0% collision rate across all evaluated adversarial congestion and multi-swarm urban scenarios.
The system eliminated persistent starvation and deadlocks, whereas learning-only, reactive, and coordination-reduced baselines exhibited navigation failures under identical conditions.