Safe-SCAD, Predicting Driver Behavior [Webpage]

2019 - Present

Sponsored by Toyota InfoTech & University of York

I'm the primary researcher in the Safe-SCAD project, a collaborative project between UVA, Carnegie Mellon University, University of York. The project aims to understand the driver’s role in the safe operation of the semi-autonomous vehicles. I contributed to the project by:

• Designing, developing and integrating various mixed traffic driving scenarios using Matlab Simulink..
• Managing and training a team of 4 undergrad research assistants primarily responsible for lab experimentation data entry, and coding.
• Developing surveys, recruting 25 participants (including pilots).
• Measuring task completion, reaction time, error rates, and quality of takeover.
• Pre-processing and analyzing user's subjective and objective data (GSR, Heart Rate, Eye-Tracker, and EEG)
• Developing the first neural network model for reliably predicting multiple aspects of driver's takeover behavior, called DeepTake, by using Python, and Tensorflow.
• Evaluating the performance of DeepTake and compared with traditional machine learning techniques(SVM, Logistic Reg., Naive Bayes, Random Forest) where DeepTake improved the performance from 78% to 93% on average.

Driver Perception and Safety Feedback

2018 - 2020

Sponsored by UVA School of Data Science

The goal of my project is developing a novel methodology to predict the cognitive states of Autonomous Vehicle drivers and alerts them such that they are able to respond in a timely and informed fashion. As the data scientist and human factors engineer I facilitated:

• Conducting literature reviews across human factors fundamentals and emerging technologies for perception and cognition applications.
• Writing grant proposals, designing, and conducting extensive study on the effect of takeover modalities on perception and mental readiness
• Developing driving scenarios using Simulink and Unity
• Measing task completion time, reaction time, error rates, and biometric data (HR, eye-Tracker).
• Analyzing EEG data using spectral power analyses with MATLAB Toolbox EEGLab and NeuroScan software.
• Analyzing surveys and physiological data from 10 individuals using univariate, multivariate, regression, and advanced statistical analyses to predict how cognitive state and mental readiness would change wrt. type of given alarm.
• Developed machine learning algorithms that consider external hazards in conjunction with drivers’ cognitive states.

Cognitive Trust in Human-Autonomous Vehicle Interactions

2018 - 2020

Sponsored by NSF & Toyota InfoTech

• Developed a signal Temporal Logic (STL) to check evolution of the driver trust to the automated system.
• Formalized the human-automation interaction as a partially observable Markov decision process (POMDP) and model trust as a partially observable state.

Effect of Worker's mood on productivity

2015 - 2016

Led the creation, execution, and analysis of individuals' mood (e.g. appetite and tiredness), emotion (e.g. happy, anxious, and distressed), and cognitive state on productivity and social contagion.

• Planned and wrote the proposal on investigating the effect of workers' mood changes on the productivity throughout a shift.
• Applied human factors principles and usability heuristics to identify the strengths and shortcomings of existing interactions in the workplace.
• Designed, ran, analyzed and presented ethnographic research.
• Led the data collection, interviews and survey responses capturing employees feedback and moods.
• Captured mood states prior and during a shift by qualitative methods (interviews and observation) and modeled their negative and positive moods in the subsequent time period using quantitative methods (parametric statistics)
• Presented the research findings, usability data, and product development process and the main effect of mood change on each phase on the Proceeding of WSC conference 2016.