Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up
tumeteor 
posted an update 18 days ago
Post
3,930
**Release: Safety-CHORES-RL**

We’ve released **Safety-CHORES-RL**, a new rollout-native simulation framework for RL-based vision-language-action (VLA) tasks, built on Safety-CHORES.

The framework supports:
- **Reset-and-replay branching:** reconstruct an episode state and explore different policy continuations.
- **Budgeted rollout search:** evaluate candidate actions and search for successful trajectories.
- **Checked trajectory execution:** retain a selected trajectory and verify state and transition consistency during execution.

The website includes animated trajectories, branches with different outcomes, episode statistics, reproducible examples, and API documentation.

🌐 [Website and demos](https://tumeteor.github.io/safety-chores-rl/)
💻 [Source code](https://github.com/tumeteor/safety-chores-rl)
📖 [API reference](https://tumeteor.github.io/safety-chores-rl/guides/api.html)
🔬 [Validation and limitations](https://tumeteor.github.io/safety-chores-rl/guides/release.html)

Thanks to the Safety-CHORES / SafeVLA team for the underlying benchmark, tasks, and safety infrastructure.

Feedback, use cases, and contributions are welcome!

The split between rollout horizon and native-step budget caught my eye: replaying a long prefix is real work even when the continuation is short. Counting that replay cost should make comparisons between planners much easier to interpret. For BestOfN versus RecedingHorizon, I'd look at both native steps and wall time alongside success, since equal rollout counts need not mean equal compute.

In this post