SEAL: Towards Safe Autonomous Driving via Skill-Enabled Adversary Learning for Closed-Loop Scenario Generation

2024-09-16

arXiv.org

AI Summary

Existing autonomous driving verification methods generate safety-critical scenarios with oversimplified objectives, leading to distorted adversarial behaviors—either overly aggressive or unresponsive—and lacking diversity and realism. To address this, we propose a skill-augmented adversarial learning framework that, for the first time, jointly learns objective functions and models human driving skills to establish a closed-loop, human-factor-driven scenario generation mechanism. Our approach integrates reinforcement learning–based skill representation, adversarial perturbation optimization, simulation-in-the-loop feedback training, and human behavioral prior modeling. Experiments demonstrate that our method improves ego-vehicle task success rates by over 20% on both in-distribution and out-of-distribution real-world scenarios, significantly outperforming state-of-the-art methods. The source code and toolchain are publicly released.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Verification and validation of autonomous driving (AD) systems and components is of increasing importance, as such technology increases in real-world prevalence. Safety-critical scenario generation is a key approach to robustify AD policies through closed-loop training. However, existing approaches for scenario generation rely on simplistic objectives, resulting in overly-aggressive or non-reactive adversarial behaviors. To generate diverse adversarial yet realistic scenarios, we propose SEAL, a scenario perturbation approach which leverages learned objective functions and adversarial, human-like skills. SEAL-perturbed scenarios are more realistic than SOTA baselines, leading to improved ego task success across real-world, in-distribution, and out-of-distribution scenarios, of more than 20%. To facilitate future research, we release our code and tools: https://github.com/cmubig/SEAL

Problem

• Enhance autonomous driving safety
• Generate realistic adversarial scenarios
• Improve ego task success rates

Innovation

• Skill-Enabled Adversary Learning
• Closed-Loop Scenario Generation
• Learned Objective Functions