aTTC-CBF: Time-based barrier framework halves drone collisions, doubles progress

A team of researchers (Benedikt Barthel Sorensen, Mitchell Black, Erfaun Noorani, and Themistoklis Sapsis) has published a new framework for collision avoidance in multi-agent autonomous aerial vehicles, available on arXiv (2608.14239). Their approach rethinks the standard control barrier function (CBF) safety layer by defining the barrier directly in time rather than distance or velocity. The core innovation is adversarial time-to-collision (aTTC), a risk metric that calculates how quickly any surrounding agent could reach a given aircraft under adversarial intent. This shifts the safety logic from reacting to current collision courses to proactively adjusting velocity based on the minimum possible collision time given each agent's dynamic constraints.

To make this practical, the team trained a differentiable neural-network surrogate that computes aTTC in real time within a standard CBF quadratic program, avoiding heavy optimization overhead. In long-horizon 3D simulations of independent-pursuit and formation-flight scenarios, the resulting aTTC-CBF framework demonstrated dramatic improvements: up to twice the waypoint progress and half the collision rate compared to a higher-order distance-based CBF baseline. This suggests that time-based safety metrics are not only more selective—allowing closer flight—but also improve mission efficiency in dynamic, uncertain, and adversarial environments. The paper is relevant to swarm robotics, autonomous logistics, and defense applications where reliable yet non-conservative collision avoidance is critical.