Graph Neural Planning and Predictive Control for Multi-Robot Communication-Constrained Unlabeled Motion Planning
Manohari Goarin, Yang Zhou, Giuseppe Loianno
AI summary
Problem
Existing GNN-based planners rely on simplified dynamics and simulation, overlooking real-world challenges like nonlinear dynamics, actuation limits, and communication constraints.
Approach
The authors propose a hierarchical architecture where a Graph Attention Planner generates intermediate subgoals via minimal 1-hop communication, tracked by a decentralized NMPC that enforces safety and dynamic constraints.
Key results
- GATP outperforms GCN baselines and generalizes to teams of up to 50 robots
- Maintains robust task performance under communication delays up to 200 ms
- Successfully deployed on 4 real-world quadrotors with decentralized on-board inference
- Ablation confirms proposed MLP fusion update function reduces assignment conflicts
Why it matters
Enables scalable, safe, and communication-robust multi-robot coordination for real-world deployment in logistics, search-and-rescue, and surveillance.
Abstract
The multi-robot unlabeled motion planning prob- lem of concurrently assigning robots to goals and generating safe trajectories is central in many collaborative tasks. Recent Graph Neural Network methods offer scalable decentralized solutions but rely on simplified dynamics and simulation envi- ronments, overlooking key challenges of real-world deployment such as dynamic feasibility and communication constraints. To address these gaps, we propose a hierarchical framework that combines a Graph ATtention Planner (GATP) with a decentral- ized Nonlinear Model Predictive Controller (NMPC). GATP provides intermediate subgoals through multi-robot coopera- tion, and the NMPC enforces safety under nonlinear dynamics and actuation constraints. We evaluate our framework in both simulation and real-world quadrotor experiments. Thanks to attention mechanisms and minimal communication require- ments, we demonstrate improved generalization to larger teams, robustness to communication delays up to 200 ms and practical feasibility with decentralized on-board inference.