Learning to Interact in Multi-Agent Self-Play for Controllable Autonomous Driving Scenario Generation
Organizations: Zoox, Foster City, USA
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.
Figures & tables
| San Francisco | Austin | |||
|---|---|---|---|---|
| With agent relative goals? | Yes | No | Yes | No |
| cutoff | 84.2% | 14.6% | 79.0% | 1.0% |
| cutoff_and_slow_down | 70.7% | 3.7% | 84.7% | 0.1% |
| leading | 98.2% | 100.0% | 99.8% | 100.0% |
| on_right | 100.0% | 81.6% | 100.0% | 72.8% |
| tailing | 100.0% | 94.8% | 99.7% | 91.5% |
Autonomous Driving Scenario Generation
Autonomous driving scenario generation creates realistic, controllable traffic situations for training and testing vehicles, especially in rare and safety-critical cases.
See all 120 papers