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

We present relay policy learning, a method for imitation and reinforcementlearning that can solve multi-stage, long-horizon robotic tasks. This generaland universally-applicable, two-phase approach consists of an imitationlearning stage that produces goal-conditioned hierarchical policies, and areinforcement learning phase that finetunes these policies for taskperformance. Our method, while not necessarily perfect at imitation learning,is very amenable to further improvement via environment interaction, allowingit to scale to challenging long-horizon tasks. We simplify the long-horizonpolicy learning problem by using a novel data-relabeling algorithm for learninggoal-conditioned hierarchical policies, where the low-level only acts for afixed number of steps, regardless of the goal achieved. While we rely ondemonstration data to bootstrap policy learning, we do not assume access todemonstrations of every specific tasks that is being solved, and insteadleverage unstructured and unsegmented demonstrations of semantically meaningfulbehaviors that are not only less burdensome to provide, but also can greatlyfacilitate further improvement using reinforcement learning. We demonstrate theeffectiveness of our method on a number of multi-stage, long-horizonmanipulation tasks in a challenging kitchen simulation environment. Videos areavailable at https://relay-policy-learning.github.io/

Relay Policy Learning: Solving Long-Horizon Tasks via Imitation and Reinforcement Learning