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

State-of-the-art reinforcement learning algorithms predominantly learn apolicy from either a numerical state vector or images. Both approachesgenerally do not take structural knowledge of the task into account, which isespecially prevalent in robotic applications and can benefit learning ifexploited. This work introduces a neural network architecture that combinesrelational inductive bias and visual feedback to learn an efficient positioncontrol policy for robotic manipulation. We derive a graph representation thatmodels the physical structure of the manipulator and combines the robot'sinternal state with a low-dimensional description of the visual scene generatedby an image encoding network. On this basis, a graph neural network trainedwith reinforcement learning predicts joint velocities to control the robot. Wefurther introduce an asymmetric approach of training the image encoderseparately from the policy using supervised learning. Experimental resultsdemonstrate that, for a 2-DoF planar robot in a geometrically simplistic 2Denvironment, a learned representation of the visual scene can replace access tothe explicit coordinates of the reaching target without compromising on thequality and sample efficiency of the policy. We further show the ability of themodel to improve sample efficiency for a 6-DoF robot arm in a visuallyrealistic 3D environment.

Graph Neural Networks for Relational Inductive Bias in Vision-based Deep Reinforcement Learning of Robot Control