DecompGAIL: Learning Realistic Traffic Behaviors with Decomposed Multi-Agent Generative Adversarial Imitation Learning

2025-10-08

AI Summary

Existing imitation learning methods struggle to model realistic traffic behavior: behavioral cloning suffers from covariate shift, while Generative Adversarial Imitation Learning (GAIL) exhibits severe training instability in multi-agent settings—primarily due to misleading discriminative signals induced by neighbor interactions. To address this, we propose DecompGAIL, a framework that decouples behavior authenticity into two orthogonal discriminative components—“ego-vehicle–map” and “ego-vehicle–neighbors”—thereby eliminating spurious interaction-induced interference. We further introduce a socialized PPO objective to enhance inter-agent behavioral consistency. Our method integrates GAIL, behavioral cloning, distance-weighted neighborhood rewards, and a lightweight SMART backbone network. Evaluated on the WOMD Sim Agents 2025 benchmark, DecompGAIL achieves state-of-the-art performance, significantly improving both simulation fidelity and training stability.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Multiagent Systems: Adversarial Agents
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Realistic traffic simulation is critical for the development of autonomous driving systems and urban mobility planning, yet existing imitation learning approaches often fail to model realistic traffic behaviors. Behavior cloning suffers from covariate shift, while Generative Adversarial Imitation Learning (GAIL) is notoriously unstable in multi-agent settings. We identify a key source of this instability: irrelevant interaction misguidance, where a discriminator penalizes an ego vehicle's realistic behavior due to unrealistic interactions among its neighbors. To address this, we propose Decomposed Multi-agent GAIL (DecompGAIL), which explicitly decomposes realism into ego-map and ego-neighbor components, filtering out misleading neighbor: neighbor and neighbor: map interactions. We further introduce a social PPO objective that augments ego rewards with distance-weighted neighborhood rewards, encouraging overall realism across agents. Integrated into a lightweight SMART-based backbone, DecompGAIL achieves state-of-the-art performance on the WOMD Sim Agents 2025 benchmark.

Problem

• Addresses unstable multi-agent imitation learning in traffic simulation
• Filters misleading neighbor interactions to improve behavior realism
• Enhances autonomous driving simulation through decomposed adversarial learning

Innovation

• Decomposes realism into ego-map and ego-neighbor components
• Introduces social PPO with distance-weighted neighborhood rewards
• Integrates decomposition into lightweight SMART-based backbone