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

Progress in deep learning highlights the tremendous potential of utilizingdiverse robotic datasets for attaining effective generalization and makes itenticing to consider leveraging broad datasets for attaining robustgeneralization in robotic learning as well. However, in practice, we often wantto learn a new skill in a new environment that is unlikely to be contained inthe prior data. Therefore we ask: how can we leverage existing diverse offlinedatasets in combination with small amounts of task-specific data to solve newtasks, while still enjoying the generalization benefits of training on largeamounts of data? In this paper, we demonstrate that end-to-end offline RL canbe an effective approach for doing this, without the need for anyrepresentation learning or vision-based pre-training. We present pre-trainingfor robots (PTR), a framework based on offline RL that attempts to effectivelylearn new tasks by combining pre-training on existing robotic datasets withrapid fine-tuning on a new task, with as few as 10 demonstrations. PTR utilizesan existing offline RL method, conservative Q-learning (CQL), but extends it toinclude several crucial design decisions that enable PTR to actually work andoutperform a variety of prior methods. To our knowledge, PTR is the first RLmethod that succeeds at learning new tasks in a new domain on a real WidowXrobot with as few as 10 task demonstrations, by effectively leveraging anexisting dataset of diverse multi-task robot data collected in a variety of toykitchens. We also demonstrate that PTR can enable effective autonomousfine-tuning and improvement in a handful of trials, without needing anydemonstrations. An accompanying overview video can be found in thesupplementary material and at thi URL: https://sites.google.com/view/ptr-final/

Pre-Training for Robots: Offline RL Enables Learning New Tasks from a Handful of Trials