Multi-Head Attention Machine Learning for Fault Classification in Mixed Autonomous and Human-Driven Vehicle Platoons

Context:

As a graduation requirement for the Engineering Science program, all students must complete a undergraduate thesis research project under supervision from a faculty professor. For my thesis project, I was co-supervised by Prof. Deepa Kundur, at the University of Toronto, and Prof. Mohammad Al Janaideh, at Memorial University (UPDATE: now at University of Guelph). In addition to the final thesis report, I was also able to contribute a paper to ICRA2023 that encompasses much of my thesis work. I was fortunate for this paper to be accepted and I travelled to London, UK to present the work as first author in a poster session.

Overview:

This work introduces a multi-head attention model, based off the transformer encoder, that is able to classify fault and abnormality behavior in mixed autonomous and human driven vehicle platoons. The model was trained on data extracted from SimuLink simulations of the platoon and validated on data collected from a real-world laboratory platoon.

Availability: The paper manuscript can be downloaded in PDF format here.

The poster presented at ICRA can be downloaded in PDF format here.

A recorded six-minute presentation video can be downloaded in MP4 format here.

The code for model training and testing was written in Python and can be found here.