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

Current Reinforcement Learning (RL) algorithms struggle with long-horizontasks where time can be wasted exploring dead ends and task progress may beeasily reversed. We develop the SPOT framework, which explores within actionsafety zones, learns about unsafe regions without exploring them, andprioritizes experiences that reverse earlier progress to learn with remarkableefficiency.  The SPOT framework successfully completes simulated trials of a variety oftasks, improving a baseline trial success rate from 13% to 100% when stacking 4cubes, from 13% to 99% when creating rows of 4 cubes, and from 84% to 95% whenclearing toys arranged in adversarial patterns. Efficiency with respect toactions per trial typically improves by 30% or more, while training takes just1-20k actions, depending on the task.  Furthermore, we demonstrate direct sim to real transfer. We are able tocreate real stacks in 100% of trials with 61% efficiency and real rows in 100%of trials with 59% efficiency by directly loading the simulation-trained modelon the real robot with no additional real-world fine-tuning. To our knowledge,this is the first instance of reinforcement learning with successful sim toreal transfer applied to long term multi-step tasks such as block-stacking androw-making with consideration of progress reversal. Code is available athttps://github.com/jhu-lcsr/good_robot .

“Good Robot!”: Efficient Reinforcement Learning for Multi-Step Visual Tasks with Sim to Real Transfer