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

Solving goal-conditioned tasks with sparse rewards using self-supervisedlearning is promising because of its simplicity and stability over currentreinforcement learning (RL) algorithms. A recent work, called Goal-ConditionedSupervised Learning (GCSL), provides a new learning framework by iterativelyrelabeling and imitating self-generated experiences. In this paper, we revisitthe theoretical property of GCSL -- optimizing a lower bound of the goalreaching objective, and extend GCSL as a novel offline goal-conditioned RLalgorithm. The proposed method is named Weighted GCSL (WGCSL), in which weintroduce an advanced compound weight consisting of three parts (1) discountedweight for goal relabeling, (2) goal-conditioned exponential advantage weight,and (3) best-advantage weight. Theoretically, WGCSL is proved to optimize anequivalent lower bound of the goal-conditioned RL objective and generatesmonotonically improved policies via an iterated scheme. The monotonic propertyholds for any behavior policies, and therefore WGCSL can be applied to bothonline and offline settings. To evaluate algorithms in the offlinegoal-conditioned RL setting, we provide a benchmark including a range of pointand simulated robot domains. Experiments in the introduced benchmarkdemonstrate that WGCSL can consistently outperform GCSL and existingstate-of-the-art offline methods in the fully offline goal-conditioned setting.

Rethinking Goal-conditioned Supervised Learning and Its Connection to Offline RL