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

Learning with sparse rewards remains a significant challenge in reinforcementlearning (RL), especially when the aim is to train a policy capable ofachieving multiple different goals. To date, the most successful approaches fordealing with multi-goal, sparse reward environments have been model-free RLalgorithms. In this work we propose PlanGAN, a model-based algorithmspecifically designed for solving multi-goal tasks in environments with sparserewards. Our method builds on the fact that any trajectory of experiencecollected by an agent contains useful information about how to achieve thegoals observed during that trajectory. We use this to train an ensemble ofconditional generative models (GANs) to generate plausible trajectories thatlead the agent from its current state towards a specified goal. We then combinethese imagined trajectories into a novel planning algorithm in order to achievethe desired goal as efficiently as possible. The performance of PlanGAN hasbeen tested on a number of robotic navigation/manipulation tasks in comparisonwith a range of model-free reinforcement learning baselines, includingHindsight Experience Replay. Our studies indicate that PlanGAN can achievecomparable performance whilst being around 4-8 times more sample efficient.

PlanGAN: Model-based Planning With Sparse Rewards and Multiple Goals