Controllable Sim Agents with Behavior Latents

2026-07-02

AI Summary

This work addresses the challenge of achieving realistic and interpretable agent behavior control in traffic simulation. The authors propose the CNeVA framework, which integrates Gaussian behavioral latent variables with a rectified flow trajectory generator, augmented by classifier-free guidance and a soft eligibility gating mechanism to effectively mitigate sparse reward issues. By introducing closed-form conjugate variational updates, mixed-channel mask curriculum learning, and a contextual residual return metric, CNeVA enables monotonic, interpretable, and reward-tamper-free control over key driving dimensions—such as speed, acceleration, and safety—for the first time in high-fidelity imitation learning. Evaluated on the Waymo Open Motion Dataset, CNeVA achieves state-of-the-art realism and demonstrates that physically grounded constraints are essential for preventing reward tampering.

Abstract

Realistic traffic simulation requires agents that imitate logged behavior and can also be steered along interpretable axes. Such controllability enables engineers to isolate variables, reproduce specific edge cases, and test autonomous systems without real-world risk. We introduce Controllable Neural Variational Agents (CNeVA), a controllable simulated-agent framework that learns to infer a per-agent Gaussian behavior latent from per-channel discounted returns via a closed-form conjugate variational update, conditioning a rectified-flow trajectory generator trained on a mixed channel-mask curriculum for classifier-free guidance. To tackle scarcity in reward signals, we propose soft eligibility gates that replace hard binary thresholds with smooth exponential decay, preserving the gradient signal for near-threshold agents. On the Waymo Open Motion Dataset, CNeVA attains competitive realism on the benchmark while exposing per-channel controllability that the higher-ranked imitation models lack. Speed- and acceleration-based steering produces monotone responses without stall-induced reward hacking. Safety controllability is monotone and substantial with the introduction of soft eligibility. We manage to achieve steerable map compliance under a context-residual return measure. Furthermore, our experiment demonstrates that steering metrics must be read alongside physical-plausibility guardrails to avoid reward-hacking confounds.

Problem

• controllable simulation
• behavior modeling
• traffic simulation
• autonomous driving
• imitation learning

Innovation

• controllable simulation
• behavior latent
• rectified flow
• soft eligibility gates
• classifier-free guidance