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

Specifying reward signals that allow agents to learn complex behaviors is along-standing challenge in reinforcement learning. A promising approach is toextract preferences for behaviors from unlabeled videos, which are widelyavailable on the internet. We present Video Prediction Rewards (VIPER), analgorithm that leverages pretrained video prediction models as action-freereward signals for reinforcement learning. Specifically, we first train anautoregressive transformer on expert videos and then use the video predictionlikelihoods as reward signals for a reinforcement learning agent. VIPER enablesexpert-level control without programmatic task rewards across a wide range ofDMC, Atari, and RLBench tasks. Moreover, generalization of the video predictionmodel allows us to derive rewards for an out-of-distribution environment whereno expert data is available, enabling cross-embodiment generalization fortabletop manipulation. We see our work as starting point for scalable rewardspecification from unlabeled videos that will benefit from the rapid advancesin generative modeling. Source code and datasets are available on the projectwebsite: https://escontrela.me/viper

Video Prediction Models as Rewards for Reinforcement Learning