I want to know how to let AI controllers fly spacecraft without breaking safety constraints. The work starts with ARCS, a simulator for testing AI and classical controllers on rendezvous with a keep-out zone.
Open-source software. In progress.
An open-source, headless simulator for testing large language model agents and classical controllers on spacecraft rendezvous with keep-out safety constraints. First result: a PD baseline violates the keep-out zone in 100% of 30 obstructed approaches.
Paper in progress.
A benchmark, built on ARCS, comparing LLM agents, classical controllers, and safety-filtered versions on spacecraft rendezvous.
Paper in progress.
Safety-shielded AI control for spacecraft approaching unknown asteroids.