sylph

SYLPH: Social Behavior as a Key to Learning-Based Multi-Agent Pathfinding Dilemmas

This repository provides the pretrained model for SYLPH, introduced in:

Social Behavior as a Key to Learning-Based Multi-Agent Pathfinding Dilemmas

SYLPH is a learning-based Multi-Agent Path Finding (MAPF) framework that explicitly models social behaviors among agents to improve decentralized coordination in challenging multi-agent interactions.

The pretrained checkpoint provided here corresponds to the policy used for evaluation in the paper.

Model Overview

Multi-Agent Path Finding requires multiple agents to navigate toward their individual goals while avoiding collisions with obstacles and other agents.

Learning-based decentralized MAPF methods can struggle in challenging interaction scenarios, particularly when agents encounter coordination dilemmas caused by competing paths and limited shared space.

SYLPH introduces social behavior into the learned policy to facilitate coordination among agents and improve their ability to resolve these interactions.

Key Features

• Decentralized multi-agent path planning
• Learning-based coordination
• Socially-aware agent behavior
• Multi-agent reinforcement learning
• Designed for challenging MAPF interaction scenarios
• Shared policy across agents

Model Files

The repository contains the pretrained SYLPH policy checkpoint:

net_checkpoint.pkl