# Steerable Adversarial Scenario Generation through Test-Time Preference Alignment

arXiv: https://arxiv.org/abs/2509.20102 · HTML: https://mlsanity.com/paper/2509.20102/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Reward modeling and RLHF, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 12 of its 50 references that are in the map.
- First posted on arXiv 24 Sep 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)

## Builds on (most cited first, 12 of 12)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [Editing Models with Task Arithmetic](https://mlsanity.com/paper/2212.04089/) (2022)
- [Git Re-Basin: Merging Models modulo Permutation Symmetries](https://mlsanity.com/paper/2209.04836/) (2022)
- [Failure-Scenario Maker for Rule-Based Agent using Multi-agent Adversarial Reinforcement Learning and its Application to Autonomous Driving](https://mlsanity.com/paper/1903.10654/) (2019)
- [Bone Soups: A Seek-and-Soup Model Merging Approach for Controllable Multi-Objective Generation](https://mlsanity.com/paper/2502.10762/) (2025)
- [FREA: Feasibility-Guided Generation of Safety-Critical Scenarios with Reasonable Adversariality](https://mlsanity.com/paper/2406.02983/) (2024)
- [Seeking to Collide: Online Safety-Critical Scenario Generation for Autonomous Driving with Retrieval Augmented Large Language Models](https://mlsanity.com/paper/2505.00972/) (2025)
- [RIFT: Group-Relative RL Fine-Tuning for Realistic and Controllable Traffic Simulation](https://mlsanity.com/paper/2505.03344/) (2025)
- [Direct Preference Optimization-Enhanced Multi-Guided Diffusion Model for Traffic Scenario Generation](https://mlsanity.com/paper/2502.12178/) (2025)
- [AED: Automatic Discovery of Effective and Diverse Vulnerabilities for Autonomous Driving Policy with Large Language Models](https://mlsanity.com/paper/2503.20804/) (2025)
- [PEO: Improving Bi-Factorial Preference Alignment with Post-Training Policy Extrapolation](https://mlsanity.com/paper/2503.01233/) (2025)

## Built on by (most cited first, 9 of 9)

- [ADV-0: Closed-Loop Min-Max Adversarial Training for Long-Tail Robustness in Autonomous Driving](https://mlsanity.com/paper/2603.15221/) (2026)
- [E3AD: An Emotion-Aware Vision-Language-Action Model for Human-Centric End-to-End Autonomous Driving](https://mlsanity.com/paper/2512.04733/) (2025)
- [World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning](https://mlsanity.com/paper/2607.10630/) (2026)
- [Sparse Threats, Focused Defense: Criticality-Aware Robust Reinforcement Learning for Safe Autonomous Driving](https://mlsanity.com/paper/2601.01800/) (2026)
- [KG-ASG: Collision-Knowledge-Guided Closed-Loop Adversarial Scenario Generation With Primary-Support Attribution](https://mlsanity.com/paper/2605.18895/) (2026)
- [EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents](https://mlsanity.com/paper/2606.03678/) (2026)
- [From Attacks to Curricula: Learnability-Guided Adversarial Training for Safe Autonomous Driving](https://mlsanity.com/paper/2606.14032/) (2026)
- [CCFM: Collision-Constrained Flow Matching for Safety-Critical Scenario Generation](https://mlsanity.com/paper/2607.04451/) (2026)
- [How corner is a corner case? Percentile control for highway scenario generation](https://mlsanity.com/paper/2610.05003/) (2026)

## Abstract (arXiv)

Adversarial scenario generation is a cost-effective approach for safety assessment of autonomous driving systems. However, existing methods are often constrained to a single, fixed trade-off between competing objectives such as adversariality and realism. This yields behavior-specific models that cannot be steered at inference time, lacking the efficiency and flexibility to generate tailored scenarios for diverse training and testing requirements. In view of this, we reframe the task of adversarial scenario generation as a multi-objective preference alignment problem and introduce a new framework named Steerable Adversarial scenario GEnerator (SAGE). SAGE enables fine-grained test-time control over the trade-off between adversariality and realism without any retraining. We first propose hierarchical group-based preference optimization, a data-efficient offline alignment method that learns to balance competing objectives by decoupling hard feasibility constraints from soft preferences. Instead of training a fixed model, SAGE fine-tunes two experts on opposing preferences and constructs a continuous spectrum of policies at inference time by linearly interpolating their weights. We provide theoretical justification for this framework through the lens of linear mode connectivity. Extensive experiments demonstrate that SAGE not only generates scenarios with a superior balance of adversariality and realism but also enables more effective closed-loop training of driving policies. Project page: https://tongnie.github.io/SAGE/.