A Practical Guide to Building a White-Box End-to-End Autonomous Driving Testbed with Open-Source AD and CAN-Based Drive-By-Wire Vehicle

Autonomous driving (AD) systems face various security threats, including attacks on sensors and machine learning modules. End-to-end (E2E) security evaluations are essential for understanding how component-level attacks translate into control-level consequences, and they are primarily conducted in simulation environments because of cost efficiency and safety concerns. However, recent studies demonstrate gaps between simulations and physical environments, underscoring the need for physical testing. Although commercial vehicles provide a potential test platform in physical experiments, their closed-source nature limits explainability and reproducibility, and integrating open-source alternatives can be challenging due to a lack of documentation.

To address these challenges, we present systematic documentation of integration challenges and practical solutions encountered when implementing open-source AD systems on hardware platforms. As a concrete implementation, we demonstrate the integration of Autoware with a CAN-based drive-by-wire system for E2E evaluation environment.

This demo shows videos of our recent efforts to build a white-box testbed that integrates Autoware with a CAN-based drive-by-wire vehicle for security research as as a concrete implementation of our practical integration guideline. The demo includes AD capabilities and E2E security evaluations of emergency stopping functions and adversarial attacks on pedestrian detection.