NASA-backed AI keeps planes from crashing mid-air in real time
AI predicts mid-air collisions and reroutes planes in under 6 seconds using real-world flight data.
A team from the University of Michigan and NASA has developed Plan-and-Avoid (PAA), an AI-powered real-time trajectory coordination system for aircraft operating in congested airspace. The framework prioritizes high-priority flights (e.g., emergency landings or critical missions) while dynamically adjusting nearby aircraft paths to prevent mid-air collisions. Using real-world ADS-B (Automatic Dependent Surveillance-Broadcast) data from Washington, D.C.'s airspace, the researchers simulated over 140 hours of flight across 900+ forced-landing scenarios.
PAA demonstrated remarkable efficiency, generating feasible advisories for all 575 unique conflict encounters with a worst-case end-to-end response time of just 5.7 seconds on a standard personal computer—including priority trajectory planning and advisory generation. Notably, 93.5% of these advisories complied with the RTCA DO-365 temporal threshold of 35 seconds for detect-and-avoid systems, underscoring its potential for real-world aviation safety applications. Future work will focus on quantifying operational impacts, such as delays introduced by advisory compliance.
- PAA prioritizes critical flights (e.g., emergencies) and dynamically reroutes nearby aircraft to avoid collisions in real time.
- Tested on 140+ hours of real-world flight data from Washington, D.C., resolving 575 unique conflicts with a 5.7s worst-case response time.
- 93.5% of generated advisories met aviation safety standards (RTCA DO-365), paving the way for AI-driven air traffic control.
Why It Matters
AI could soon automate high-stakes air traffic decisions, reducing human workload and improving flight safety in dense airspace.