CARE Drive A Framework for Evaluating Reason-Responsiveness of Vision Language Models in Automated Driving
arXiv:2602.15645v1 Announce Type: new Abstract: Foundation models, including vision language models, are increasingly used in automated driving to interpret scenes, recommend actions, and generate natural language explanations. However, existing evaluation methods primarily assess outcome based performance, such as safety and trajectory accuracy, without determining whether model decisions reflect human relevant considerations. As a result, it remains unclear whether explanations produced by such models correspond to genuine reason responsive decision making or merely post hoc rationalizations. This limitation is especially significant in safety critical domains because it can create false confidence. To address this gap, we propose CARE Drive, Context Aware Reasons Evaluation for Driving, a model agnostic framework for evaluating reason responsiveness in vision language models applied to automated driving. CARE Drive compares baseline and reason augmented model decisions under contro
arXiv:2602.15645v1 Announce Type: new Abstract: Foundation models, including vision language models, are increasingly used in automated driving to interpret scenes, recommend actions, and generate natural language explanations. However, existing evaluation methods primarily assess outcome based performance, such as safety and trajectory accuracy, without determining whether model decisions reflect human relevant considerations. As a result, it remains unclear whether explanations produced by such models correspond to genuine reason responsive decision making or merely post hoc rationalizations. This limitation is especially significant in safety critical domains because it can create false confidence. To address this gap, we propose CARE Drive, Context Aware Reasons Evaluation for Driving, a model agnostic framework for evaluating reason responsiveness in vision language models applied to automated driving. CARE Drive compares baseline and reason augmented model decisions under controlled contextual variation to assess whether human reasons causally influence decision behavior. The framework employs a two stage evaluation process. Prompt calibration ensures stable outputs. Systematic contextual perturbation then measures decision sensitivity to human reasons such as safety margins, social pressure, and efficiency constraints. We demonstrate CARE Drive in a cyclist overtaking scenario involving competing normative considerations. Results show that explicit human reasons significantly influence model decisions, improving alignment with expert recommended behavior. However, responsiveness varies across contextual factors, indicating uneven sensitivity to different types of reasons. These findings provide empirical evidence that reason responsiveness in foundation models can be systematically evaluated without modifying model parameters.
Executive Summary
This article proposes CARE Drive, a model-agnostic framework for evaluating reason-responsiveness in vision language models applied to automated driving. The framework assesses whether human reasons causally influence model decisions, addressing the limitation of existing evaluation methods that primarily focus on outcome-based performance. CARE Drive's two-stage evaluation process involves prompt calibration and systematic contextual perturbation to measure decision sensitivity to human reasons. The authors demonstrate the framework's effectiveness in a cyclist overtaking scenario, showing that explicit human reasons significantly influence model decisions and improve alignment with expert-recommended behavior. The findings provide empirical evidence that reason responsiveness in foundation models can be systematically evaluated without modifying model parameters, highlighting the potential of CARE Drive to enhance the safety and trustworthiness of automated driving systems.
Key Points
- ▸ CARE Drive is a model-agnostic framework for evaluating reason-responsiveness in vision language models
- ▸ The framework assesses whether human reasons causally influence model decisions
- ▸ CARE Drive's two-stage evaluation process involves prompt calibration and systematic contextual perturbation
Merits
Systematic evaluation of reason-responsiveness
CARE Drive provides a structured approach to evaluating the reason-responsiveness of vision language models, addressing the limitation of existing evaluation methods.
Model-agnostic framework
The framework's model-agnostic design allows it to be applied to various vision language models, making it a versatile tool for evaluating reason-responsiveness.
Improved safety and trustworthiness
By evaluating the reason-responsiveness of vision language models, CARE Drive has the potential to enhance the safety and trustworthiness of automated driving systems.
Demerits
Complexity of implementation
The framework's two-stage evaluation process may be complex to implement, requiring significant computational resources and expertise.
Limited generalizability
The framework's effectiveness may be limited to specific scenarios or domains, requiring further validation and adaptation for broader applicability.
Expert Commentary
The proposal of CARE Drive is a timely and significant contribution to the field of automated driving, addressing a critical limitation of existing evaluation methods. The framework's systematic evaluation of reason-responsiveness provides a much-needed structured approach to assessing the decision-making processes of vision language models. However, the complexity of implementation and limited generalizability of the framework's effectiveness are significant challenges that require careful consideration. Furthermore, the findings of this study have significant implications for the development and deployment of autonomous vehicles, highlighting the need for more effective and transparent AI systems. As the field continues to evolve, it is essential to prioritize the development of frameworks like CARE Drive that can provide insights into the decision-making processes of AI systems.
Recommendations
- ✓ Further research is needed to address the complexity of implementation and limited generalizability of the framework's effectiveness
- ✓ Regulatory agencies should consider adopting CARE Drive or similar frameworks to evaluate the reason-responsiveness of vision language models in automated driving systems
