PCASim: Promptable Closed-Loop Adversarial Simulation for Urban Traffic Environment
Chuancheng Zhang, Zhenhao Wang, Kaizheng Li, Yaran Lin, Qiang Guo, Bin Jiang
AI summary
Problem
Traditional autonomous driving testing relies on inefficient real-world data or isolated scenario generation methods that fail to integrate data, knowledge, and adversarial approaches, lacking a closed-loop system for co-evolving scenario generation and agent training.
Approach
The authors construct a dynamic adversarial scenario repository from real traffic data and use a retrieval-augmented LLM to convert user prompts into executable simulation code, then train reinforcement learning agents to navigate these scenarios in a continuous feedback loop that enriches the dataset.
Key results
- Improved domain-specific language generation accuracy by 12%
- Increased newly generated scenario transformation success rate by 8%
- Enhanced obstacle-avoidance capability by 30%
- Constructed a dynamic, retrieval-augmented adversarial scenario repository
Why it matters
It provides autonomous driving developers with a scalable, automated pipeline for generating and validating safety-critical corner cases, accelerating robust model training and reducing reliance on costly real-world testing.
Abstract
Real-world autonomous driving, particularly in urban environments with numerous corner cases, requires rigorous testing to ensure product safety and robustness. How- ever, few studies have explored integrating adversarial scenario generation with the training of safety agents in closed-loop testing, enabling efficient co-evolution and mutual enhancement of both. To address this challenge, an adversarial behavior knowledge repository is constructed by applying rule-based filtering to an open-source dataset, combined with knowledge retrieval modules tailored for simulation environments. A large language model (LLM) is employed to integrate knowledge-, data-, and adversarial-driven approaches, generating safety- critical traffic scenarios customized to user needs. Additionally, while evaluating the generated scenarios, we employ rein- forcement learning models to train the behaviors of different types of vehicles, thereby enriching scenario diversity beyond existing datasets while preserving realism. Experimental results demonstrate that the proposed framework improves the accuracy of domain-specific language generation by 12%. Moreover, the success rate of newly generated scenario transformations increases by 8%, while obstacle-avoidance capability is en- hanced by 30%. For the complete manuscript, please refer to: https://zhenhaooo.github.io/PCASim.github.io/