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

Imitation learning is a promising approach to help robots acquire dexterousmanipulation capabilities without the need for a carefully-designed reward or asignificant computational effort. However, existing imitation learningapproaches require sophisticated data collection infrastructure and struggle togeneralize beyond the training distribution. One way to address this limitationis to gather additional data that better represents the full operatingconditions. In this work, we investigate characteristics of such additionaldemonstrations and their impact on performance. Specifically, we study theeffects of corrective and randomly-sampled additional demonstrations onlearning a policy that guides a five-fingered robot hand through apick-and-place task. Our results suggest that corrective demonstrationsconsiderably outperform randomly-sampled demonstrations, when the proportion ofadditional demonstrations sampled from the full task distribution is largerthan the number of original demonstrations sampled from a restrictive trainingdistribution. Conversely, when the number of original demonstrations are higherthan that of additional demonstrations, we find no significant differencesbetween corrective and randomly-sampled additional demonstrations. Theseresults provide insights into the inherent trade-off between the effortrequired to collect corrective demonstrations and their relative benefits overrandomly-sampled demonstrations. Additionally, we show that inexpensivevision-based sensors, such as LeapMotion, can be used to dramatically reducethe cost of providing demonstrations for dexterous manipulation tasks. Our codeis available athttps://github.com/GT-STAR-Lab/corrective-demos-dexterous-manipulation.

Evaluating the Effectiveness of Corrective Demonstrations and a Low-Cost Sensor for Dexterous Manipulation