ViSafe: A Vision-Only System for High-Speed Airborne Collision Avoidance
In the rapidly evolving world of autonomous aerial systems, ensuring safe separation between airborne vehicles is a critical challenge—especially as uncrewed aerial systems (UAS) increasingly share airspace with traditional aircraft. Current collision avoidance systems often rely on active sensors like transponders, radar, or ADS-B, which are impractical for small UAS due to size, weight, power, and cost (SWaP-C) constraints. Enter ViSafe, a groundbreaking vision-only airborne collision avoidance system developed by researchers at Carnegie Mellon University. This system leverages edge AI and multi-camera hardware to enable high-speed detect-and-avoid (DAA) capabilities without relying on external communication or bulky sensors.
How ViSafe Works
ViSafe builds on the team’s prior work, AirTrack, a deep learning-based aircraft detection and tracking system. The system integrates a custom multi-camera payload with a real-time edge AI framework, processing high-resolution video feeds to detect and track intruder aircraft. By using Control Barrier Functions (CBFs)—a mathematical tool for ensuring safety in dynamic systems—ViSafe provides provable runtime guarantees for collision avoidance, even at closure rates of up to 144 km/h (40 m/s).
Key components of ViSafe include:
- Multi-Camera Detection & Tracking – AirTrack’s neural network processes synchronized camera feeds to detect and track intruders, even at long ranges.
- 3D State Estimation – A Kalman filter fuses detections across multiple views to estimate intruder position and velocity.
- Safety-Critical Control – A CBF-based supervisory controller modifies the drone’s flight path in real time to maintain safe separation.
Real-World and Simulated Testing
The team conducted extensive testing, including 80+ hours of real-world flight tests and 12,000 simulated trajectories using a high-fidelity digital twin environment. ViSafe successfully avoided collisions in diverse scenarios, including head-on, overtaking, and lateral encounters, even under challenging lighting and weather conditions.
Why This Matters
ViSafe represents a major leap forward in autonomous aerial safety, particularly for small UAS operating in congested airspace. By relying solely on vision—a lightweight, cost-effective alternative to radar or transponders—it opens the door to scalable, high-density drone operations without compromising safety. The system’s ability to handle high-speed encounters sets a new benchmark for vision-only collision avoidance, paving the way for future regulatory approvals and commercial adoption.
For more details, check out the full paper on arXiv.