On Learning Closed-Loop Probabilistic Multi-Agent Simulator

TL;DR

This paper presents NIVA, a probabilistic hierarchical Bayesian framework for closed-loop multi-agent traffic simulation that unifies trajectory prediction and simulation models, enabling realistic, diverse, and controllable autonomous vehicle scenario generation.

Contribution

NIVA introduces a novel Bayesian approach to multi-agent simulation that unifies trajectory prediction and closed-loop simulation with enhanced control over agent behaviors.

Findings

NIVA achieves competitive performance on the Waymo dataset.

NIVA provides improved control over agent intentions and driving styles.

The framework unifies existing models under a Bayesian inference perspective.

Abstract

The rapid iteration of autonomous vehicle (AV) deployments leads to increasing needs for building realistic and scalable multi-agent traffic simulators for efficient evaluation. Recent advances in this area focus on closed-loop simulators that enable generating diverse and interactive scenarios. This paper introduces Neural Interactive Agents (NIVA), a probabilistic framework for multi-agent simulation driven by a hierarchical Bayesian model that enables closed-loop, observation-conditioned simulation through autoregressive sampling from a latent, finite mixture of Gaussian distributions. We demonstrate how NIVA unifies preexisting sequence-to-sequence trajectory prediction models and emerging closed-loop simulation models trained on Next-token Prediction (NTP) from a Bayesian inference perspective. Experiments on the Waymo Open Motion Dataset demonstrate that NIVA attains competitive performance compared to the existing method while providing embellishing control over intentions and driving styles.