Home
Tempest is a shield synthesis tool for probabilistic environments. tempestpy provides Python bindings for Tempest, making it straightforward to integrate shielding into reinforcement learning workflows via Gymnasium-compatible wrappers.
See Installation to get started and Usage for a guide to the main concepts of tempestpy. The Examples section shows shielded RL training in practice, including the RL training library MinigridSafe — an extension of Minigrid that adds uncertainty and adversarial actors to grid-world environments. The full API documentation is available under API Reference.
Tempest and tempestpy are maintained by the Trusted AI group at Graz University of Technology. You can reach us via mail.
News
2025-05-20 — TODO