Answers are grounded in this paper and cite their sources.
This paper develops a safety-gated graph-attention compensation backstepping architecture for multi-UAV formation flight among static and dynamic obstacles. Each follower is represented by a three-dimensional double-integrator model with bounded acceleration. A sliding-surface law tracks the desired formation; an obstacle-distance authority factor and a safety-conflict projection coordinate tracking and avoidance; a graph attention network (GAT) generates risk-aware additive compensation; and a safety gate redistributes authority among tracking, obstacle avoidance, inter-UAV anti-collision, altitude constraint, and compensation channels. Smooth saturation enforces the input bound. A Lyapunov analysis of the smooth nominal control law establishes uniform ultimate boundedness in the unsaturated operating region, while restart-enabled covariance matrix adaptation evolution strategy (CMA-ES) tunes the ten coupled gains offline. In a dense dynamic scene, the tuned controller attains 90.0% obstacle-avoidance success, 66.7% joint success, a mean minimum obstacle clearance of 0.249±0.020\,m, and a tracking RMSE of $0.879$\,m. Removing GAT compensation increases RMSE to $2.40$\,m, and CMA-ES tuning increases dense-scene joint success from $3.3%$ to $66.7%$. Ablations identify obstacle avoidance and predictive avoidance as safety-critical modules, nominal tracking and GAT compensation as performance-critical modules, and the residual inter-UAV margin as the dominant remaining bottleneck.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: