Effects of Unreliable Level of Automation and Takeover Time Budget on Young Drivers’ Takeover Performance And Workload

Funding: U.S. department of transportation

My Role: graduate leader and researcher

Timeline: Aug 2019_ Dec 2020

Research Method: Experiment design, eye-tracking, Statistical hypothesis, human-computer interaction, surveys, cognitive

How did I come up with the idea?

Young drivers:

Prone to distraction
More likely to be engaged in non-driving related tasks (NDRT)
High number of crashes compared to other age groups

Highly automated driving (HAD):

Supports the driver with longitudinal and lateral vehicle control
Let drivers to be engaged in an NDRT
Might need an occasional intervene of the driver

Research objectives:

1. Assess the effect of automation on young drivers’ takeover performance and workload

2. Find the most efficient takeover time budget (TOTB) to improve young drivers’ takeover performance and reduce workload (5s, 8s, or 10s)

Takeover process

How did I come up with the idea?

Young drivers:

Prone to distraction
More likely to be engaged in non-driving related tasks (NDRT)
High number of crashes compared to other age groups

Highly automated driving (HAD):

Supports the driver with longitudinal and lateral vehicle control
Let drivers to be engaged in an NDRT
Might need an occasional intervene of the driver

Research hypotheses:

Driving simulation study

Apparatus:

A fixed-base driving simulator (Real time technologies)
Eye-tracker (Pupil-core)

Participants:

Twenty-eight young drivers (14 males and 14 females)
Age: 25.14 ± 3.34 yrs (got from national highway traffic safety administration)
20/20 vision

Design of experiment

Within-subject design (2 Driving conditions × 3 TOTBs)
Two critical incidents (i.e., a lead vehicle that braked suddenly) in each scenario
Drivers were asked to perform a non-driving related task when driving in one of the critical incidents
Visual-auditory takeover request
Randomized order to avoid learning effect

Variables

Driving activity load index (DALI) questionnaire

Modification of the NASA-TLX questionnaire
A measure of the perceived workload of drivers
Six subscales

Data collection/ Procedure

Informed consent form/ Demographic questionnaire/ Covid forms (10 min)
Eye-tracking Calibration/ Baseline measurement (10 min)
Driving simulator training scenario/ Data analysis (10 min)
NDRT training/ Driving a scenario with the NDRT (10 min)
Six driving scenarios (each 6 min)/ DALI and rest after each scenario (4 min)/ (50 min)
Simulator sickness questionnaire (SSQ) after each two trials

Data analysis

Data screening to identify outliers
Diagnostics (normality and variance check)
In case of parametric assumption violations:
Data transformation was used
Ordinal logistic regression analysis: To analyze the secondary task accuracy
Covariates: Age, gender, experience in automated driving studies, experience with automated vehicles, and trial number (1-6)

Findings: TOTB (Takeover performance)

Note: Letters on figures resulted from post-hoc analysis. Columns with different letters are significantly different. For example, 5 s of TOTB leads to a significantly shorter reaction time as compared to 8s and 10s.

Findings: TOTB (Workload)

Findings: Highly automated driving (HAD)

Impact on Industry: Design recommendations for vehicle manufacturing companies

When designing highly autonomous vehicles for young drivers (i.e., age between 16- and 30-year-old):

1. Do not use 5 seconds (or less) of takeover time budget:

Low-quality (unsafe) takeover performance
High workload
Potential loss of customer
Reduce customer trust in your brand

2. If you are limited on time and budget, use 8 seconds of takeover time budget:

Safe takeover performance
Higher workload than manual mode
Use training modules on handling takeover situations to build customer trust and reduce workload

3. If you are not limited on time and budget, use 10 seconds of takeover time budget:

Safe takeover performance
Low workload

Impact on Industry: Design recommendations for vehicle manufacturing companies

Journal publication:

Shahini, F., Park, J., & Zahabi, M. Effects of Unreliable Automation, Non-Driving Related Task, and Takeover 2 Time Budget on Drivers’ Takeover Performance and Cognitive Workload. Ergonomics, 1-14.

Conference presentation:

Shahini, F., Park, J., & Zahabi, M. (2021, September). Effects of unreliable automation and takeover time budget on young drivers’ mental workload. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting (Vol. 65, No. 1, pp. 1082-1086). Sage CA: Los Angeles, CA: SAGE Publications.

Picture of me presenting the conference paper in HFES 2021

What would I do if I had a chance to redo the experiment?

1_ Using a trust questionnaire

Trust plays an important role in human interaction with automated systems!

2_ Using a high-fidelity driving simulator

It increases the validity and credibility of data.
It is more realistic.
It includes motion.
360 degree

3_ Using a visual-haptic takeover request rather than a visual-auditory

It does not intervene with visual-auditory NDRT
I could measure the difference between visual-auditory and visual-haptic TORs.

4_ Recruiting participants from different areas

Wide variety of driving behavior

5_ Increasing the duration of scenarios

It helps to better see the possible “out-of-the-loop” effects of automation
It increases the validity of the data.

6_ Using different time of the day

All scenarios were built in day-light condition which is not generalizable to night-time

7_ Measuring the physical demand associated with each level

Despite the possible negative “out-of-the-loop” effects of automation, it can significantly reduce the physical load!
Check the trade-off between physical load, cognitive load, performance, and trust!