By Cooper Wang Class of 2025

Detection of Black Ice in Autonomous Vehicles Using Inertial Measurement Based Binary Classification Neural Network

Editor’s Note: The Taft STEM Research Fellowship is a yearlong, advanced study program for students pursuing interdisciplinary STEM research beyond the classroom. Fellows work closely with faculty and expert mentors, collaborate with peers, and apply their research to real-world problems, culminating in a public presentation to faculty and field professionals. The course combines independent scholarship with structured support and offers opportunities to explore research that bridges multiple STEM disciplines.

Abstract

Black ice poses significant challenges to driving, specifically autonomous driving, due to its difficulty to detect and its impacts on vehicle safety. Present methods for detecting black ice, although accurate, are still vulnerable to external environmental influences and cannot function in certain environments. Therefore, the research looks into novel methods of all environment black ice detection, using inertial measurement data collected with a scale model of vehicles to train neural networks for binary classification of road conditions. The resulting method from two separate neural network structures are 98.8% and 99.5% accurate respectively, and deployment of the neural network onto Tensor Processing Units (TPU) is proved to be feasible with the average inference time being 0.75 milliseconds and the standard deviation being 0.13 milliseconds. A Two Proportions Z-Test also proves the method’s improvement in accuracy to be statistically significant.

Poster

TO LEARN MORE, CLICK HERE TO VIEW PDF POSTER.