# VILTA: A VLM-in-the-Loop Adversary for Enhancing Driving Policy Robustness

arXiv: https://arxiv.org/abs/2601.12672 · HTML: https://mlsanity.com/paper/2601.12672/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Adversarial scenarios and safety testing, one of 9 directions in End-to-end autonomous driving.
- Builds on 26 of its 48 references that are in the map.
- First posted on arXiv 19 Jan 2026.

## Placed in

- End-to-end autonomous driving / [Adversarial scenarios and safety testing](https://mlsanity.com/driving/adversarial-scenarios-and-safety-testing/)

## Builds on (most cited first, 26 of 26)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor](https://mlsanity.com/paper/1801.01290/) (2018)
- [nuScenes: A multimodal dataset for autonomous driving](https://mlsanity.com/paper/1903.11027/) (2019)
- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://mlsanity.com/paper/2507.06261/) (2025)
- [RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control](https://mlsanity.com/paper/2307.15818/) (2023)
- [CARLA: An Open Urban Driving Simulator](https://mlsanity.com/paper/1711.03938/) (2017)
- [SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations](https://mlsanity.com/paper/2108.01073/) (2021)
- [Behavioral Cloning from Observation](https://mlsanity.com/paper/1805.01954/) (2018)
- [End-to-end Autonomous Driving: Challenges and Frontiers](https://mlsanity.com/paper/2306.16927/) (2023)
- [DenseTNT: End-to-end Trajectory Prediction from Dense Goal Sets](https://mlsanity.com/paper/2108.09640/) (2021)
- [MetaDrive: Composing Diverse Driving Scenarios for Generalizable Reinforcement Learning](https://mlsanity.com/paper/2109.12674/) (2021)
- [Evolving Curricula with Regret-Based Environment Design](https://mlsanity.com/paper/2203.01302/) (2022)
- [AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles](https://mlsanity.com/paper/2101.06549/) (2021)
- [ChatScene: Knowledge-Enabled Safety-Critical Scenario Generation for Autonomous Vehicles](https://mlsanity.com/paper/2405.14062/) (2024)
- ["Curse of rarity" for autonomous vehicles](https://mlsanity.com/paper/2207.02749/) (2022)
- [CAT: Closed-loop Adversarial Training for Safe End-to-End Driving](https://mlsanity.com/paper/2310.12432/) (2023)
- [Versatile Behavior Diffusion for Generalized Traffic Agent Simulation](https://mlsanity.com/paper/2404.02524/) (2024)
- [Scenario Dreamer: Vectorized Latent Diffusion for Generating Driving Simulation Environments](https://mlsanity.com/paper/2503.22496/) (2025)
- [Training Adversarial Agents to Exploit Weaknesses in Deep Control Policies](https://mlsanity.com/paper/2002.12078/) (2020)
- [Causal Composition Diffusion Model for Closed-loop Traffic Generation](https://mlsanity.com/paper/2412.17920/) (2024)
- [CaDRE: Controllable and Diverse Generation of Safety-Critical Driving Scenarios using Real-World Trajectories](https://mlsanity.com/paper/2403.13208/) (2024)
- [From Words to Collisions: LLM-Guided Evaluation and Adversarial Generation of Safety-Critical Driving Scenarios](https://mlsanity.com/paper/2502.02145/) (2025)
- [DriveGen: Towards Infinite Diverse Traffic Scenarios with Large Models](https://mlsanity.com/paper/2503.05808/) (2025)
- [CrashAgent: Crash Scenario Generation via Multi-modal Reasoning](https://mlsanity.com/paper/2505.18341/) (2025)
- [Stackelberg Driver Model for Continual Policy Improvement in Scenario-Based Closed-Loop Autonomous Driving](https://mlsanity.com/paper/2309.14235/) (2023)

## Built on by (most cited first, 6 of 6)

- [Xiaomi OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation](https://mlsanity.com/paper/2604.18486/) (2026)
- [Toward Physically Consistent Driving Video World Models under Challenging Trajectories](https://mlsanity.com/paper/2603.24506/) (2026)
- [DriveReward: A Comprehensive Dataset and Generative Vision-Language Reward Model for Autonomous Driving](https://mlsanity.com/paper/2606.08525/) (2026)
- [DriveVer: Lightweight Trajectory Evaluator as Test-Time Verifier for Autonomous Driving](https://mlsanity.com/paper/2607.00399/) (2026)
- [Latent-Centroid Steering: Single-Pass Classifier-Free Guidance for Command-Aligned Autonomous Driving](https://mlsanity.com/paper/2608.00237/) (2026)
- [One Policy, Many Embodiments: Unified Camera-Centric Action Geometry Pre-training for Heterogeneous Embodied Manipulation](https://mlsanity.com/paper/2608.26058/) (2026)

## Abstract (arXiv)

The safe deployment of autonomous driving (AD) systems is fundamentally hindered by the long-tail problem, where rare yet critical driving scenarios are severely underrepresented in real-world data. Existing solutions including safety-critical scenario generation and closed-loop learning often rely on rule-based heuristics, resampling methods and generative models learned from offline datasets, limiting their ability to produce diverse and novel challenges. While recent works leverage Vision Language Models (VLMs) to produce scene descriptions that guide a separate, downstream model in generating hazardous trajectories for agents, such two-stage framework constrains the generative potential of VLMs, as the diversity of the final trajectories is ultimately limited by the generalization ceiling of the downstream algorithm. To overcome these limitations, we introduce VILTA (VLM-In-the-Loop Trajectory Adversary), a novel framework that integrates a VLM into the closed-loop training of AD agents. Unlike prior works, VILTA actively participates in the training loop by comprehending the dynamic driving environment and strategically generating challenging scenarios through direct, fine-grained editing of surrounding agents' future trajectories. This direct-editing approach fully leverages the VLM's powerful generalization capabilities to create a diverse curriculum of plausible yet challenging scenarios that extend beyond the scope of traditional methods. We demonstrate that our approach substantially enhances the safety and robustness of the resulting AD policy, particularly in its ability to navigate critical long-tail events.