Most modern vehicles are not merely mechanical machines but they have increasingly become embedded systems with Electronic Control Unit (ECU), an array of sensors, and computerized control. CAN bus, a standard protocol specifies how vehicle’s micro-controllers and other ECUs communication with each other and with an onboard sensor to exchange vehicle data, primarily designed for diagnostics and maintenance purposes. Recently, interest has grown to exploit the CAN Bus interface to analyze high-throughput on-board vehicle data in an attempt to understand driving behavior and analyze driving data aiming a multitude of goals such as studying driving patterns, security vulnerabilities, and employing smart vehicles for enhanced human-driving assistance. In this talk, I present our research on a similar subject where we exploit the CAN Bus mechanism of a few modern vehicles with a commercially available CAN-to-USB hardware, equipped with in-house developed software libraries and data-analytic tools. I present our findings including challenges in terms of decoding vendor-specific CAN messages, characterizing data-throughput, and assessing the quality of CAN Data. I briefly describe the software solution libpanda and an open-source data analysis tool strym for downstream analysis of captured CAN bus data.
Back to Mathematical Challenges and Opportunities for Autonomous Vehicles