AI Safety & Governance · Recorded milestone
Runtime safety bounds for a learning flight controller
Recorded year2024
Physical UAV experiments combine a reinforcement-learning controller with a runtime-assurance layer enforcing specified position and velocity constraints.
Explore this milestone on the interactive mapBuilt on
Led to
Sources & evidence
- Demonstrating Reinforcement Learning and Run Time Assurance for Spacecraft Inspection Using Unmanned Aerial Vehicles 2024-05-10
Physical UAV tests combine a reinforcement-learning controller with runtime assurance enforcing position and velocity constraints while emulating spacecraft inspection; not an on-orbit or universal-safety demonstration.
- A Verification Framework for Runtime Assurance of Autonomous UAS 2024
Formal reasoning about runtime-assurance monitors and specified safety properties.
- NASA autonomous systems and Multi-Monitor Run-Time Assurance
Runtime monitoring and interventions for bounded autonomous-system safety; no claim of universal AI alignment.
Dunlap and colleagues
Record reviewed 2026-10-18