Multi-agent system creates targeted driving scenarios for testing

Learning to Interact in Multi-Agent Self-Play for Controllable Autonomous Driving Scenario Generation

RoboticsMachine Learning

Summary

Testing self-driving cars requires simulating tricky situations where cars interact in risky ways, like tailgating or cutting off others. The authors found that existing simulation methods focus on fixed map goals and can't easily create these complex interactions. They developed a new approach that lets simulated cars pursue goals relative to other cars, enabling more realistic and controllable scenarios. This helps test autonomous vehicles better by exposing them to rare but important driving behaviors.

What this means in practice

  • For autonomous vehicle developers: Generate targeted and challenging driving scenarios in simulation to rigorously test AV driving policies against rare risky behaviors.
  • For automotive safety teams: Create controllable interactive driving cases to improve evaluation protocols for vehicle safety systems and driver assistance technologies.

Authors

Beyazit Yalcinkaya, Ethan Pronovost, Maximilian Naumann, Marcell Vazquez-Chanlatte

Abstract

Large-scale, multi-agent self-play trains robust driving policies effectively in simulation, but controlling these models to generate specific, safety-critical edge cases remains a key challenge. Existing goal-conditioned self-play frameworks rely primarily on static map-location goals and fail to reliably produce complex agent interactions such as aggressive tailgating or adversarial cut-offs. These agent interactions are necessary for rigorous autonomous vehicle (AV) testing. To address this, we present an extended multi-agent self-play framework that introduces agent-relative dynamic goals. By conditioning policies on this extended goal representation, our framework enables targeted, controllable scenario synthesis. We demonstrate that our approach successfully generates interactive driving behaviors that static-goal baselines cannot simulate, providing a flexible tool for long-tail scenario generation in AV testing.