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

Robot learning holds the promise of learning policies that generalizebroadly. However, such generalization requires sufficiently diverse datasets ofthe task of interest, which can be prohibitively expensive to collect. In otherfields, such as computer vision, it is common to utilize shared, reusabledatasets, such as ImageNet, to overcome this challenge, but this has provendifficult in robotics. In this paper, we ask: what would it take to enablepractical data reuse in robotics for end-to-end skill learning? We hypothesizethat the key is to use datasets with multiple tasks and multiple domains, suchthat a new user that wants to train their robot to perform a new task in a newdomain can include this dataset in their training process and benefit fromcross-task and cross-domain generalization. To evaluate this hypothesis, wecollect a large multi-domain and multi-task dataset, with 7,200 demonstrationsconstituting 71 tasks across 10 environments, and empirically study how thisdata can improve the learning of new tasks in new environments. We find thatjointly training with the proposed dataset and 50 demonstrations of anever-before-seen task in a new domain on average leads to a 2x improvement insuccess rate compared to using target domain data alone. We also find that datafor only a few tasks in a new domain can bridge the domain gap and make itpossible for a robot to perform a variety of prior tasks that were only seen inother domains. These results suggest that reusing diverse multi-task andmulti-domain datasets, including our open-source dataset, may pave the way forbroader robot generalization, eliminating the need to re-collect data for eachnew robot learning project.

Bridge Data: Boosting Generalization of Robotic Skills with Cross-Domain Datasets