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

Offline meta-reinforcement learning (meta-RL) methods, which adapt to unseentarget tasks with prior experience, are essential in robot control tasks.Current methods typically utilize task contexts and skills as prior experience,where task contexts are related to the information within each task and skillsrepresent a set of temporally extended actions for solving subtasks. However,these methods still suffer from limited performance when adapting to unseentarget tasks, mainly because the learned prior experience lacks generalization,i.e., they are unable to extract effective prior experience from meta-trainingtasks by exploration and learning of continuous latent spaces. We propose aframework called decoupled meta-reinforcement learning (DCMRL), which (1)contrastively restricts the learning of task contexts through pulling insimilar task contexts within the same task and pushing away different taskcontexts of different tasks, and (2) utilizes a Gaussian quantizationvariational autoencoder (GQ-VAE) for clustering the Gaussian distributions ofthe task contexts and skills respectively, and decoupling the exploration andlearning processes of their spaces. These cluster centers which serve asrepresentative and discrete distributions of task context and skill are storedin task context codebook and skill codebook, respectively. DCMRL can acquiregeneralizable prior experience and achieve effective adaptation to unseentarget tasks during the meta-testing phase. Experiments in the navigation androbot manipulation continuous control tasks show that DCMRL is more effectivethan previous meta-RL methods with more generalizable prior experience.

Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills