English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Training-time safety violations have been a major concern when we deployreinforcement learning algorithms in the real world. This paper explores thepossibility of safe RL algorithms with zero training-time safety violations inthe challenging setting where we are only given a safe but trivial-rewardinitial policy without any prior knowledge of the dynamics model and additionaloffline data. We propose an algorithm, Co-trained Barrier Certificate for SafeRL (CRABS), which iteratively learns barrier certificates, dynamics models, andpolicies. The barrier certificates, learned via adversarial training, ensurethe policy's safety assuming calibrated learned dynamics model. We also add aregularization term to encourage larger certified regions to enable betterexploration. Empirical simulations show that zero safety violations are alreadychallenging for a suite of simple environments with only 2-4 dimensional statespace, especially if high-reward policies have to visit regions near the safetyboundary. Prior methods require hundreds of violations to achieve decentrewards on these tasks, whereas our proposed algorithms incur zero violations.

Learning Barrier Certificates: Towards Safe Reinforcement Learning with Zero Training-time Violations