Validation of a Pedestrian Simulator for interaction between Pedestrians and Autonomous Vehicles

Validation of a Pedestrian Simulator for interaction between Pedestrians and Autonomous Vehicles

Agarwal, Rakshit (TU Delft Mechanical, Maritime and Materials Engineering)

Happee, R. (mentor)
de Winter, J.C.F. (mentor)
Eisma, Y.B. (graduation committee)

Delft University of Technology

Mechanical Engineering | Vehicle Engineering

2019-07-09

A large number of pedestrian fatalities and the emergence of autonomous vehicles (AVs) creates a need for research in vehicle to pedestrian (V2P) interaction. Gestures (e.g., hand movement, eye contact) used for communication between pedestrians and manually driven vehicle plays a significant role to reduce ambiguities and make safe road crossing decisions. With autonomous vehicles on the roads in the near future, it may be possible that pedestrians are not able to understand the vehicle behaviour or vice a versa. Virtual reality (VR) simulations are used to test a new method of communications such as external human-machine interface (eHMI) between AVs and pedestrians. Virtual reality has emerged as an effective and cost-efficient tool in various fields such as surgical training, technician training and driver simulators. Among many benefits of virtual reality, is it’s ability to expose its users to critical situations where the real environment would be dangerous and therefore, could be a viable tool for conducting V2P research. However, it is essential to understand how pedestrians’ behaviour and perception match in VR or 360◦ video with real life to rely on them
as an experimental method. The main goal of this study was to examine the
effectiveness of VR and 360-degree video environments in terms of pedestrian perception and behaviour to draw actionable conclusions on the reliability of VR.
An experiment was conducted to test the interaction between pedestrians and autonomous vehicles using three different environments: a field test on a a public road, a replication of the test in an animated virtual reality (VR) simulation and a third repetition using a 360-degree video recording of the same location. The animated virtual environment was designed using Unity and C#, and the 360-degree videos were recorded at the field test location from participants’ position at a height of 180 cm. The participant was standing on the sidewalk on a two-lane road in Delft in each environment. To compare human perception of distance and speed and behaviour in terms of gap accepted during road crossing scenarios in 360-degree video and animated VR, three tasks being: 1) Distance estimation, 2) Speed estimation and 3) Gap acceptance was conducted in each environment with a within-subject design (N = 30).

Results showed that there were no significant differences between the three environments for distance estimation, gap acceptance tasks and measures of presence. However, significant differences were found between animated VR and 360-degree video environment for perceived speed of 30 km/h. According to the median values, the perceived distance in real life and animated virtual environment deviated below 5% from the ground truth values. However, the median values for perceived distance in 360-degree video environment for
distances of 40 m and 63 m were overestimated. The deviation in perceived distance increased as the ground truth value increased, wherein the animated virtual environment showed the highest amount of variance amongst the three environments. Underestimation of the speed of about 5 - 10 km/h was observed in both field tests and animated virtual environment for speeds below 50 km/h. The median values show that the perceived speed was most accurate in a 360-degree video environment. However, the animated virtual environment was more consistent with real life perceived speed. A significant positive correlation between distance and speed estimates showed consistency in human perception across the three environments. A negative correlation was found between perceived distance and gap acceptance ratio, whereas a positive correlation was found between perceived speed and gap acceptance ratio. The presence scores were relatively higher for field tests and 360-degree video environments compared to the animated virtual environment. However, the scores were high enough to validate the three environments subjectively. Only one participant experience motion
sickness in virtual reality tasks with a score of above 2 out of 3, experiencing a moderate sickness but was successfully able to complete the whole experimental task. Overall results show that pedestrian perception and behaviour in 360-degree video and animated VR environment doesn’t differ significantly from real life. These results have a strong implication on the usage of virtual environments for studying pedestrians’ perception and behaviour when interacting with vehicles.

vehicle to pedestrian communication
Virtual Reality
Automated driving
Validation

http://resolver.tudelft.nl/uuid:e1581f7c-aa7e-4606-991b-bbaf7940c79c

Student theses

master thesis

© 2019 Rakshit Agarwal