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

This work shows that value-aware model learning, known for its numeroustheoretical benefits, is also practically viable for solving challengingcontinuous control tasks in prevalent model-based reinforcement learningalgorithms. First, we derive a novel value-aware model learning objective bybounding the model-advantage i.e. model performance difference, between twoMDPs or models given a fixed policy, achieving superior performance to priorvalue-aware objectives in most continuous control environments. Second, weidentify the issue of stale value estimates in naively substituting value-awareobjectives in place of maximum-likelihood in dyna-style model-based RLalgorithms. Our proposed remedy to this issue bridges the long-standing gap intheory and practice of value-aware model learning by enabling successfuldeployment of all value-aware objectives in solving several continuous controlrobotic manipulation and locomotion tasks. Our results are obtained withminimal modifications to two popular and open-source model-based RL algorithms-- SLBO and MBPO, without tuning any existing hyper-parameters, while alsodemonstrating better performance of value-aware objectives than these baselinein some environments.

Model-Advantage and Value-Aware Models for Model-Based Reinforcement Learning: Bridging the Gap in Theory and Practice