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 Tools annual

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 monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Collecting and automatically obtaining reward signals from real roboticvisual data for the purposes of training reinforcement learning algorithms canbe quite challenging and time-consuming. Methods for utilizing unlabeled datacan have a huge potential to further accelerate robotic learning. We considerhere the problem of performing manipulation tasks from pixels. In such tasks,choosing an appropriate state representation is crucial for planning andcontrol. This is even more relevant with real images where noise, occlusionsand resolution affect the accuracy and reliability of state estimation. In thiswork, we learn a latent state representation implicitly with deep reinforcementlearning in simulation, and then adapt it to the real domain using unlabeledreal robot data. We propose to do so by optimizing sequence-based selfsupervised objectives. These exploit the temporal nature of robot experience,and can be common in both the simulated and real domains, without assuming anyalignment of underlying states in simulated and unlabeled real images. Wepropose Contrastive Forward Dynamics loss, which combines dynamics modellearning with time-contrastive techniques. The learned state representationthat results from our methods can be used to robustly solve a manipulation taskin simulation and to successfully transfer the learned skill on a real system.We demonstrate the effectiveness of our approaches by training a vision-basedreinforcement learning agent for cube stacking. Agents trained with our method,using only 5 hours of unlabeled real robot data for adaptation, shows a clearimprovement over domain randomization, and standard visual domain adaptationtechniques for sim-to-real transfer.

Self-Supervised Sim-to-Real Adaptation for Visual Robotic Manipulation