Research Analyzer
← Back ICRA 2026

Learning Policies for Dynamic Coalition Formation in Multi-Robot Task Allocation

Lucas Bezerra, Ataide Santos, Shinkyu Park

PDF

AI summary

Key figure (auto-extracted from paper)
A decentralized reinforcement learning framework enables robots to dynamically form and revise coalitions using only local information, outperforming traditional market-based methods.
Multi-Robot Task Allocation Dynamic Coalition Formation Multi-Agent Reinforcement Learning Decentralized Control MAPPO Partial Observability

Problem

Decentralized multi-robot task allocation in dynamic environments lacks efficient methods for forming coalitions under partial observability without relying on high-frequency communication.

Approach

The method extends MAPPO by integrating spatial action maps, intention sharing, and proactive task revision, allowing robots to learn collaborative policies from local observations and neighbor plans.

Key results

  • Outperforms the PCFA market-based baseline in dynamic coalition tasks
  • Scales efficiently to environments with up to 1000 robots
  • Demonstrates strong generalizability across varying task densities and levels
  • Enables proactive coalition formation and task revision using only local information

Why it matters

Provides a scalable, communication-efficient coordination strategy critical for large-scale autonomous systems in disaster response and logistics.

Abstract

We propose a decentralized, learning-based frame- work for dynamic coalition formation in Multi-Robot Task Al- location (MRTA). Our approach extends MAPPO by integrating spatial action maps, robot motion planning, intention sharing, and task allocation revision to enable effective and adaptive coalition formation. Extensive simulation studies confirm the effectiveness of our model, enabling each robot to rely solely on local information to learn timely revisions of task selections and form coalitions with other robots to complete collaborative tasks. The results also highlight the proposed framework’s ability to handle large robot populations and adapt to scenarios with diverse task sets.

Index terms

Multi-Robot Systems Cooperating Robots Reinforcement Learning

Related papers