SPACeR: Self-Play Anchoring with Centralized Reference Models

TL;DR

SPACeR introduces a scalable self-play framework for autonomous vehicle simulation that uses a pretrained tokenized model as a reference, achieving human-like behavior with faster inference and smaller models.

Contribution

The paper presents SPACeR, a novel method combining self-play RL with a pretrained tokenized reference model to improve scalability and realism in autonomous vehicle simulation.

Findings

Achieves competitive performance with imitation learning methods.

Up to 10x faster inference than large generative models.

50x smaller in parameter size compared to existing models.

Abstract

Developing autonomous vehicles (AVs) requires not only safety and efficiency, but also realistic, human-like behaviors that are socially aware and predictable. Achieving this requires sim agent policies that are human-like, fast, and scalable in multi-agent settings. Recent progress in imitation learning with large diffusion-based or tokenized models has shown that behaviors can be captured directly from human driving data, producing realistic policies. However, these models are computationally expensive, slow during inference, and struggle to adapt in reactive, closed-loop scenarios. In contrast, self-play reinforcement learning (RL) scales efficiently and naturally captures multi-agent interactions, but it often relies on heuristics and reward shaping, and the resulting policies can diverge from human norms. We propose SPACeR, a framework that leverages a pretrained tokenized autoregressive motion model as a centralized reference policy to guide decentralized self-play. The reference model provides likelihood rewards and KL divergence, anchoring policies to the human driving distribution while preserving RL scalability. Evaluated on the Waymo Sim Agents Challenge, our method achieves competitive performance with imitation-learned policies while being up to 10x faster at inference and 50x smaller in parameter size than large generative models. In addition, we demonstrate in closed-loop ego planning evaluation tasks that our sim agents can effectively measure planner quality with fast and scalable traffic simulation, establishing a new paradigm for testing autonomous driving policies.