A Bayesian inference-based feedback on car-following behavior: Improving the validity of traffic simulations with human driving input

A Bayesian inference-based feedback on car-following behavior: Improving the validity of traffic simulations with human driving input

Molenaar, R.B.

Mechanical, Maritime and Materials Engineering

Delft Center for Systems and Control

2014-03-12

Traffic simulators are useful tools that can be used for entertainment, driver education, and research into human driving behavior and the effects of innovative traffic solutions. A traffic simulator provides a high degree of control, as every traffic situation can be recreated with exactly the same conditions for all test subjects. The downside of working with a simulator is the limited validity. When simulating a new situation, observing valid driving behavior from the test subjects requires valid simulated background traffic. In order to generate valid background traffic, realistic driving models are needed for that specific situation. This creates a vicious circle, as the realistic observations that are not observed yet are required to generate realistic driving models. The purpose of this research is to generate realistic simulated traffic for new situations. Using the Bayesian inference method in an online feedback loop, observed driving behavior from test subjects is used to calibrate existing driving models. These models are fed back to the simulator, where they will be used to generate traffic with the information gathered from the subjects. The influence of implementing a Bayesian-inference based feedback loop on subjective validity of the simulated traffic is tested in various experiments. The results show that a direct comparison, where the updated traffic and non-updated traffic are observed in the same session, results in an increase in subjective validity. An indirect comparison does not lead to an increase in subjective validity. The conclusion from the results is that, although the feedback loop leads to an increase in subjective validity, the effect is not strong enough to be noticeable when comparing traffic from two separate sessions.

Bayesian inference
car-following
simulation validity

http://resolver.tudelft.nl/uuid:4cdeca5c-e83b-4bb1-b6ce-42fae7412cc2

Student theses

master thesis

(c) 2014 Molenaar, R.B.