On Learning Closed-Loop Probabilistic Multi-Agent Simulator

2025-08-01

AI Summary

This work addresses the lack of realistic, scalable, and closed-loop multi-agent traffic simulators for autonomous driving evaluation. We propose Neural Interactive Agents (NIVA), a unified framework that jointly models trajectory prediction and closed-loop simulation. NIVA employs Bayesian inference to enable fine-grained, probabilistic control over driver intent and style; it leverages a hierarchical Bayesian model coupled with latent Gaussian mixture–driven autoregressive sampling to generate diverse, highly interactive traffic scenarios conditioned on observations. Evaluated on the Waymo Open Motion Dataset, NIVA achieves prediction accuracy comparable to state-of-the-art methods while significantly improving scenario controllability, diversity, and scalability. By providing high-fidelity, editable, and physics-aware simulation, NIVA establishes a robust foundation for closed-loop evaluation of autonomous driving systems.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Multiagent Systems: Adversarial Agents
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Agentic search

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.

Problem

• Developing realistic multi-agent traffic simulators for AV evaluation
• Creating closed-loop probabilistic simulation with interactive scenarios
• Unifying trajectory prediction and simulation models via Bayesian inference

Innovation

• Hierarchical Bayesian model for multi-agent simulation
• Autoregressive sampling from Gaussian mixture distributions
• Unifies trajectory prediction and closed-loop simulation models