A Novel Hybrid XAI Solution for Autonomous Vehicles: Real-Time Interpretability Through LIME-SHAP Integration

H Ahmed Tahir ¹, Walaa Alayed ², Waqar Ul Hassan ³

⁰School of Computing, Engineering and Mathematics, Western Sydney University, Penrith, NSW 2751, Australia.

Sensors (basel, Switzerland) +

|

November 9, 2024

View abstract on PubMed

Summary

This study introduces a hybrid explainable AI (XAI) framework for autonomous vehicles (AVs), combining LIME and SHAP for transparent AI decision-making. The novel approach enhances model interpretability and efficiency for real-time AV applications.

Area Of Science

Artificial Intelligence
Computer Vision
Robotics

Background

Advancements in autonomous vehicles (AVs) and artificial intelligence (AI) necessitate transparent decision-making processes.
Existing explainable AI (XAI) methods present tradeoffs between precision, global understanding, and computational efficiency.
There is a critical need for robust XAI solutions suitable for onboard deployment in safety-critical AV systems.

Purpose Of The Study

To propose and evaluate a novel hybrid explainable AI (XAI) framework for autonomous vehicles (AVs).
To combine the strengths of Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) for enhanced transparency and efficiency.
To provide a balanced approach for onboard deployment in safety-critical AV applications.

Main Methods

A hybrid XAI framework integrating LIME and SHAP was developed.
The framework was evaluated on state-of-the-art models: ResNet-18, ResNet-50, and SegNet-50.
Performance was assessed using the KITTI dataset, focusing on fidelity, interpretability, and consistency metrics.

Main Results

The hybrid XAI framework achieved a fidelity rate exceeding 85%, an interpretability factor over 80%, and consistency above 70%.
Inference times were recorded as 0.28s (ResNet-18), 0.571s (ResNet-50), and 3.889s (SegNet), demonstrating suitability for onboard computation.
The proposed approach consistently outperformed traditional XAI methods in key performance indicators.

Conclusions

The hybrid LIME-SHAP framework offers a balanced solution for XAI in AVs, optimizing transparency and computational performance.
This research provides a strong foundation for deploying explainable AI in safety-critical autonomous driving systems.
The developed framework facilitates real-time decision-making by addressing the tradeoffs between model precision and interpretability.

Keywords:

AI/ML AV XAI self-driving vehicles unmanned AV