MAGNNET: Multi-Agent Graph Neural Network-based Efficient Task Allocation for Autonomous Vehicles with Deep Reinforcement Learning

TL;DR

MAGNNET tackles decentralized task allocation in heterogeneous, communication-constrained multi-agent systems by integrating graph neural networks with a centralized critic during training and decentralized execution. The framework uses a CTDE MARL approach with PPO updates, guided by a GNN that builds relational embeddings for agents to inform task requests while avoiding conflicts. Empirical results show a 92.5% conflict-free assignment rate and a near-centralized optimum with only a 7.49% gap from the Hungarian baseline, while scaling to up to 20 agents with an allocation time of about 2.8 seconds. The method combines A*-based path planning and a reservation-based collision resolution to ensure efficient, conflict-aware routing in a 3D grid and demonstrates robustness to dynamically generated tasks, highlighting strong practical potential for real-world autonomous vehicle coordination.

Abstract

This paper addresses the challenge of decentralized task allocation within heterogeneous multi-agent systems operating under communication constraints. We introduce a novel framework that integrates graph neural networks (GNNs) with a centralized training and decentralized execution (CTDE) paradigm, further enhanced by a tailored Proximal Policy Optimization (PPO) algorithm for multi-agent deep reinforcement learning (MARL). Our approach enables unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) to dynamically allocate tasks efficiently without necessitating central coordination in a 3D grid environment. The framework minimizes total travel time while simultaneously avoiding conflicts in task assignments. For the cost calculation and routing, we employ reservation-based A* and R* path planners. Experimental results revealed that our method achieves a high 92.5% conflict-free success rate, with only a 7.49% performance gap compared to the centralized Hungarian method, while outperforming the heuristic decentralized baseline based on greedy approach. Additionally, the framework exhibits scalability with up to 20 agents with allocation processing of 2.8 s and robustness in responding to dynamically generated tasks, underscoring its potential for real-world applications in complex multi-agent scenarios.

Figures (5)

• Figure 1: Multi-vehicle task allocation scenario with heterogeneous swarm.
• Figure 2: Multi-agent reinforcement learning architecture for task allocation.
• Figure 3: Simulation environment in PyBullet. Agents navigate with dynamic task assignments and obstacles.
• Figure 4: Mean reward vs. training steps during learning.
• Figure 5: Entropy vs. training steps during learning.