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

Safety is essential for deploying Deep Reinforcement Learning (DRL)algorithms in real-world scenarios. Recently, verification approaches have beenproposed to allow quantifying the number of violations of a DRL policy overinput-output relationships, called properties. However, such properties arehard-coded and require task-level knowledge, making their applicationintractable in challenging safety-critical tasks. To this end, we introduce theCollection and Refinement of Online Properties (CROP) framework to designproperties at training time. CROP employs a cost signal to identify unsafeinteractions and use them to shape safety properties. Hence, we propose arefinement strategy to combine properties that model similar unsafeinteractions. Our evaluation compares the benefits of computing the number ofviolations using standard hard-coded properties and the ones generated withCROP. We evaluate our approach in several robotic mapless navigation tasks anddemonstrate that the violation metric computed with CROP allows higher returnsand lower violations over previous Safe DRL approaches.

Online Safety Property Collection and Refinement for Safe Deep Reinforcement Learning in Mapless Navigation