Open-source tools for checking robot commands and studying policy behavior in simulation.
The projects explore two questions: how proposed actions can be checked before execution, and how a policy's behavior changes under controlled conditions.
- Haetae checks proposed robot actions against a policy and returns an allow, speed-cap, or deny decision. It includes a Rust core, a signed-input enforcement reference, ROS 2/Gazebo simulation, and verifiable incident logs. Start with the simulator alpha guide.
- Ataxia evaluates policy behavior offline. It compares induced behavior against a baseline using the same seeds, currently with scripted policies in a deterministic PseudoBot gridworld. See the quickstart.
Haetae is a simulator evaluation alpha and is not safety-rated. Physical-robot deployment, stopping behavior, and collision safety remain unverified. Ataxia's real-simulator and VLA adapters are planned.
Reproducible simulation results, setup reports, and documentation improvements are welcome in each repository's issues. For Haetae vulnerability reports, follow the security policy.