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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

This paper is concerned with learning transferable forward models for pushmanipulation that can be applying to novel contexts and how to improve thequality of prediction when critical information is available. We propose tolearn a parametric internal model for push interactions that, similar forhumans, enables a robot to predict the outcome of a physical interaction evenin novel contexts. Given a desired push action, humans are capable to identifywhere to place their finger on a new object so to produce a predictable motionof the object. We achieve the same behaviour by factorising the learning intotwo parts. First, we learn a set of local contact models to represent thegeometrical relations between the robot pusher, the object, and theenvironment. Then we learn a set of parametric local motion models to predicthow these contacts change throughout a push. The set of contact and motionmodels represent our internal model. By adjusting the shapes of thedistributions over the physical parameters, we modify the internal model'sresponse. Uniform distributions yield to coarse estimates when no informationis available about the novel context (i.e. unbiased predictor). A more accuratepredictor can be learned for a specific environment/object pair (e.g. lowfriction/high mass), i.e. biased predictor. The effectiveness of our approachis shown in a simulated environment in which a Pioneer 3-DX robot needs topredict a push outcome for a novel object, and we provide a proof of concept ona real robot. We train on 2 objects (a cube and a cylinder) for a total of24,000 pushes in various conditions, and test on 6 objects encompassing avariety of shapes, sizes, and physical parameters for a total of 14,400predicted push outcomes. Our results show that both biased and unbiasedpredictors can reliably produce predictions in line with the outcomes of acarefully tuned physics simulator.

Learning Transferable Push Manipulation Skills in Novel Contexts