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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Operating in the real-world often requires agents to learn about a complexenvironment and apply this understanding to achieve a breadth of goals. Thisproblem, known as goal-conditioned reinforcement learning (GCRL), becomesespecially challenging for long-horizon goals. Current methods have tackledthis problem by augmenting goal-conditioned policies with graph-based planningalgorithms. However, they struggle to scale to large, high-dimensional statespaces and assume access to exploration mechanisms for efficiently collectingtraining data. In this work, we introduce Successor Feature Landmarks (SFL), aframework for exploring large, high-dimensional environments so as to obtain apolicy that is proficient for any goal. SFL leverages the ability of successorfeatures (SF) to capture transition dynamics, using it to drive exploration byestimating state-novelty and to enable high-level planning by abstracting thestate-space as a non-parametric landmark-based graph. We further exploit SF todirectly compute a goal-conditioned policy for inter-landmark traversal, whichwe use to execute plans to "frontier" landmarks at the edge of the exploredstate space. We show in our experiments on MiniGrid and ViZDoom that SFLenables efficient exploration of large, high-dimensional state spaces andoutperforms state-of-the-art baselines on long-horizon GCRL tasks.

Successor Feature Landmarks for Long-Horizon Goal-Conditioned Reinforcement Learning