AbstractThis paper presents BIT-MX3 v0.3, a simulation study of multi-failure resilience in swarm formation control. We test four controllers (BIT-MX3 adaptive polygon, Shrink, Shrink-Adaptive, Reynolds) under Markov failure injection across 20–30 random seeds. The central finding, discovered through necessity and sufficiency ablation tests, is counterintuitive: collision avoidance in formation-based controllers is driven by polygon target-spacing geometry, not by repulsion forces, isolation logic, or damping. Adding polygon targets to a Reynolds cohesion controller reduces collisions by 100% (250→0). Increasing formation stiffness k reduces collisions monotonically. Repulsion, isolation, and damping each contribute negligibly when tested independently. Critical failure fraction f* (survival ≥ 50%) did not differentiate controllers in the configurations tested; failure rate was the dominant variable. The collision metric was the primary differentiating signal. Limitations: All results are from illustrative 2D numerical simulation (N=8 agents, Python). No hardware validation. No real UAV or robotics testing. Results should not be applied to real systems without further validation.
Bùi Quang Trịnh (Tue,) studied this question.