CollisionGAT: Controller-Agnostic One-Step Collision Screening for Multi-Agent Motion

2026-09-26

AI Summary

This study addresses the inefficiency and limited generalizability of single-step collision detection in multi-agent motion by proposing a controller-agnostic collision screening framework based on Graph Attention Networks (GATs). The approach leverages geometrically exact verification to generate supervisory labels for training the GAT model, enabling rapid single-step collision risk scoring. This framework is deeply integrated with the D* Lite planner to support continuous path following and dynamic updates of the planning graph. Experimental evaluations and independent audits demonstrate that the proposed method efficiently identifies collision risks for mobile agents under arbitrary controllers, significantly enhancing decision-making safety and system generalization capabilities.

Abstract

Before a team of robots moves, each proposed step must be checked for collisions with other robots and with obstacles. We present CollisionGAT, a graph-attention network that reads the current and proposed states of moving agents together with locally relevant stationary obstacles and returns one collision-risk score per moving agent. Any controller can use these scores to accept, repair, replan, or postpone a proposed step. We mount CollisionGAT on a continuous path-following controller and on GATeD, an obstacle-blind D* Lite planner that uses typed vetoes to update its planning graphs. Exact geometric checks supply the training labels and independently audit every executed step.

Problem

• multi-agent motion
• collision screening
• collision risk
• obstacle avoidance

Innovation

• Graph Attention Network
• Collision Screening
• Multi-Agent Motion
• Controller-Agnostic
• Path Planning