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

Model-based reinforcement learning (RL) has shown great potential in variouscontrol tasks in terms of both sample-efficiency and final performance.However, learning a generalizable dynamics model robust to changes in dynamicsremains a challenge since the target transition dynamics follow a multi-modaldistribution. In this paper, we present a new model-based RL algorithm, coinedtrajectory-wise multiple choice learning, that learns a multi-headed dynamicsmodel for dynamics generalization. The main idea is updating the most accurateprediction head to specialize each head in certain environments with similardynamics, i.e., clustering environments. Moreover, we incorporate contextlearning, which encodes dynamics-specific information from past experiencesinto the context latent vector, enabling the model to perform online adaptationto unseen environments. Finally, to utilize the specialized prediction headsmore effectively, we propose an adaptive planning method, which selects themost accurate prediction head over a recent experience. Our method exhibitssuperior zero-shot generalization performance across a variety of controltasks, compared to state-of-the-art RL methods. Source code and videos areavailable at https://sites.google.com/view/trajectory-mcl.

Trajectory-wise Multiple Choice Learning for Dynamics Generalization in Reinforcement Learning