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

Trading off performance guarantees in favor of scalability, the Multi-AgentPath Finding (MAPF) community has recently started to embrace Multi-AgentReinforcement Learning (MARL), where agents learn to collaboratively generateindividual, collision-free (but often suboptimal) paths. Scalability is usuallyachieved by assuming a local field of view (FOV) around the agents, helpingscale to arbitrary world sizes. However, this assumption significantly limitsthe amount of information available to the agents, making it difficult for themto enact the type of joint maneuvers needed in denser MAPF tasks. In thispaper, we propose SCRIMP, where agents learn individual policies from even verysmall (down to 3x3) FOVs, by relying on a highly-scalable global/localcommunication mechanism based on a modified transformer. We further equipagents with a state-value-based tie-breaking strategy to further improveperformance in symmetric situations, and introduce intrinsic rewards toencourage exploration while mitigating the long-term credit assignment problem.Empirical evaluations on a set of experiments indicate that SCRIMP can achievehigher performance with improved scalability compared to other state-of-the-artlearning-based MAPF planners with larger FOVs, and even yields similarperformance as a classical centralized planner in many cases. Ablation studiesfurther validate the effectiveness of our proposed techniques. Finally, we showthat our trained model can be directly implemented on real robots for onlineMAPF through high-fidelity simulations in gazebo.

SCRIMP: Scalable Communication for Reinforcement- and Imitation-Learning-Based Multi-Agent Pathfinding