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

We investigate whether self-supervised learning (SSL) can improve onlinereinforcement learning (RL) from pixels. We extend the contrastivereinforcement learning framework (e.g., CURL) that jointly optimizes SSL and RLlosses and conduct an extensive amount of experiments with variousself-supervised losses. Our observations suggest that the existing SSLframework for RL fails to bring meaningful improvement over the baselines onlytaking advantage of image augmentation when the same amount of data andaugmentation is used. We further perform evolutionary searches to find theoptimal combination of multiple self-supervised losses for RL, but find thateven such a loss combination fails to meaningfully outperform the methods thatonly utilize carefully designed image augmentations. After evaluating theseapproaches together in multiple different environments including a real-worldrobot environment, we confirm that no single self-supervised loss or imageaugmentation method can dominate all environments and that the currentframework for joint optimization of SSL and RL is limited. Finally, we conductthe ablation study on multiple factors and demonstrate the properties ofrepresentations learned with different approaches.

Does Self-supervised Learning Really Improve Reinforcement Learning from Pixels?