Vision language driving models struggle to follow passenger instructions
Language2Drive: Do Vision-Language-Action Models Really Follow the Instructions in Autonomous Driving?
RoboticsComputer Vision and Pattern Recognition
Summary
Self-driving car models are designed to follow spoken or written instructions while driving. This paper shows that many current vision-language-action models often drive safely but do not reliably follow different instructions when given the same road scene. The authors created a new test called Language2Drive to see if models change their behavior correctly when instructions change. They found that most models tend to repeat one driving behavior instead of adjusting to the passenger’s words, revealing a big gap in instruction-following ability.
What this means in practice
- •For autonomous vehicle engineers: Evaluate and improve driving models to better respond to varied natural-language instructions using counterfactual paired testing frameworks.
- •For simulation platform developers: Integrate counterfactual instruction-following benchmarks like Language2Drive into autonomous driving simulators to test language-driven behavior robustness.
Authors
Zihan You, Zehua Fan, Wenqi Lyu, Aoqi Wang, Ni Wang, Yifei Yang, Linge Zhao, Huan Li, Sining Ang, Yuguang Yang, Jueyu Chen, Xiangyu Li, Qianshan Wei, Jieyuan Pei, Yuxin Wang, Zhiwen Chen, Jiaqi Li, Wenxuan Song, Yanyuan Qiao, Benjin Zhu, Haoang Li, Yan Wang
Abstract
Recent Vision-Language-Action (VLA) models increasingly use natural-language instructions to condition autonomous driving behavior. Yet existing closed-loop benchmarks primarily measure route completion, safety, and rule compliance, leaving a fundamental question unresolved: does language actually determine which feasible behavior the vehicle executes? A trajectory may be safe and successful while still violating or even ignoring the passenger's instruction. To address this, we introduce \ltd, a benchmark designed to directly test this capability. Its key evaluation unit, the paired counterfactual unit (PCU), holds the initial driving scene fixed while changing only the instruction. This design enables us to distinguish isolated instruction success from counterfactual paired success across four instruction-following scopes. To test whether this distinction persists across evaluation regimes, we instantiate Language2Drive in complementary closed-loop CARLA simulation, reconstructed real-world scenes, and open-loop driving benchmarks, including our released CARLA-Language2Drive simulator with purpose-built maps. Evaluating representative driving VLAs across these settings reveals a striking mismatch between strong driving competence and weak language control: 93.2% of successful rollouts do not have a successful counterpart under the alternative feasible instruction. This failure is dominated by one-sided completion, where a model repeatedly executes one preferred behavior but fails when the same scene requires the counterfactual alternative. These results expose two fundamental limitations of current evaluation: isolated success and driving scores, which are widely used by conventional driving benchmarks, can substantially overstate language-following capability, while current driving VLAs themselves lack reliable instruction following under matched counterfactual tests.