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

Witnessing the impressive achievements of pre-training techniques onlarge-scale data in the field of computer vision and natural languageprocessing, we wonder whether this idea could be adapted in a grab-and-gospirit, and mitigate the sample inefficiency problem for visuomotor driving.Given the highly dynamic and variant nature of the input, the visuomotordriving task inherently lacks view and translation invariance, and the visualinput contains massive irrelevant information for decision making, resulting inpredominant pre-training approaches from general vision less suitable for theautonomous driving task. To this end, we propose PPGeo (Policy Pre-training viaGeometric modeling), an intuitive and straightforward fully self-supervisedframework curated for the policy pretraining in visuomotor driving. We aim atlearning policy representations as a powerful abstraction by modeling 3Dgeometric scenes on large-scale unlabeled and uncalibrated YouTube drivingvideos. The proposed PPGeo is performed in two stages to support effectiveself-supervised training. In the first stage, the geometric modeling frameworkgenerates pose and depth predictions simultaneously, with two consecutiveframes as input. In the second stage, the visual encoder learns driving policyrepresentation by predicting the future ego-motion and optimizing with thephotometric error based on current visual observation only. As such, thepre-trained visual encoder is equipped with rich driving policy relatedrepresentations and thereby competent for multiple visuomotor driving tasks.Extensive experiments covering a wide span of challenging scenarios havedemonstrated the superiority of our proposed approach, where improvements rangefrom 2% to even over 100% with very limited data.

Policy Pre-training for Autonomous Driving via Self-supervised Geometric Modeling