ERI: GRAPHSEC: Graph-Based Vehicular Communication Security with Adaptive Embedded Learning

University Of Maryland Baltimore County, Baltimore MD

Investigators

Abstract

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2). It was more than three decades ago, at the 1986 Society of Automotive Engineers (SAE) conference in Detroit, that Robert Bosch GmbH (Bosch) officially released the controller area network (CAN) specification, yet still today, this widely used in-vehicle network protocol remains as a critical security concern in modern vehicles. While automotive manufacturers race to introduce new autonomous vehicles, a society eager to deploy autonomous technologies must, with equal interest, pursue complementary efforts in securing underlying communication fabrics and gain insights from exploring existing deployed technologies. Recent studies show that cyber attacks can compromise CAN communication. This is not surprising given that vehicular communication lacks well-defined source and destination addresses within packets on which security policies may have been built and relies on the good behavior of all devices to enable functionality defined by overlaid sequences of many brief messages. These characteristics result in an attacker's single point of entry to monitor messages and broadcast unverifiable information. This research aims to improve intra-vehicular communication security and help designers integrate network properties with the sensors' physical properties to build highly robust, low-cost security systems. The proposed techniques have the potential to improve vehicular network safety and reduce cost. The PI will also conduct several educational and outreach efforts such as: (1) supervise one Ph.D. theses and several undergraduate senior design projects; (2) introduce a new course on intra-vehicular communication and security in the computer engineering curriculum at UMBC; (3) support low socioeconomic status students, through an REU program, since about 22% of the population in greater Baltimore area lives below the poverty line; (4) continue recruiting undergraduate and K-12 researchers through the local schools. In this proposal, PI first builds graphs from the CAN messages and proposes two unique approaches to secure CAN communications. The PI will apply graph-theory-based machine learning and statistical algorithms to a CAN bus to improve the reliability of the CAN bus. The intent is to develop security monitors that can be tailored to any CAN-based control system using an FPGA-based, off-the-shelf deployable platform. Notably, the PI proposes to investigate (1) Graph-based low-cost naive Bayes algorithms for vehicular security; (2) A statistical filter (SF) in the input stage of our novel scaled graph-based neural networks (GNN) for a low-cost and adaptive CAN communication; (3) Developing a flexible and configurable CAN protocol Testbed to help researchers develop new physical-property (i.e., voltage, skew, sensors & interconnect aging)-based algorithms, collect data and analyze their CAN systems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.