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

The distributional perspective on reinforcement learning (RL) has given riseto a series of successful Q-learning algorithms, resulting in state-of-the-artperformance in arcade game environments. However, it has not yet been analyzedhow these findings from a discrete setting translate to complex practicalapplications characterized by noisy, high dimensional and continuousstate-action spaces. In this work, we propose Quantile QT-Opt (Q2-Opt), adistributional variant of the recently introduced distributed Q-learningalgorithm for continuous domains, and examine its behaviour in a series ofsimulated and real vision-based robotic grasping tasks. The absence of an actorin Q2-Opt allows us to directly draw a parallel to the previous discreteexperiments in the literature without the additional complexities induced by anactor-critic architecture. We demonstrate that Q2-Opt achieves a superiorvision-based object grasping success rate, while also being more sampleefficient. The distributional formulation also allows us to experiment withvarious risk distortion metrics that give us an indication of how robots canconcretely manage risk in practice using a Deep RL control policy. As anadditional contribution, we perform batch RL experiments in our virtualenvironment and compare them with the latest findings from discrete settings.Surprisingly, we find that the previous batch RL findings from the literatureobtained on arcade game environments do not generalise to our setup.

Quantile QT-Opt for Risk-Aware Vision-Based Robotic Grasping