Intern / Master Student (Comparison of Data-Driven and Analytical Human Driving Behavior Models)
Area-Interlinking Design Analysis
4 September 2019
Open until filled
Project Title: Comparison of Data-Driven and Analytical Human Driving Behavior Models
About TUMCREATE TUMCREATE is a research platform for the improvement of Singapore's public transportation, including the deployment of electric and autonomous mobility. Researchers from Technical University Munich and Nanyang Technological University join forces and are funded by Singapore’s National Research Foundation as part of the Campus for Research Excellence and Technological Enterprise (CREATE).
In TUMCREATE, over 100 scientists, researchers and engineers work together, led by Professors from the Technical University of Munich and Nanyang Technological University. The mission of TUMCREATE is to seek the ultimate public transport system for the people of Singapore. Our innovative road transport solutions will provide high comfort and a positive travel experience, best protection of the environment and maximum benefit to the society and the economy.
Background Driving behaviour models are an integral part of agent-based traffic simulations which define the behaviour of vehicles on the road and thus need to represent reality as closely as possible. Models of human drivers become even more relevant nowadays in the context of mixed traffic simulation (humans interacting with autonomous vehicles). Typically, driving behaviour models (stochastic or deterministic) are constructed using domain knowledge and are thus analytical in nature. With recent advancements of machine learning techniques and specifically black-box, data-driven modelling approaches such as ANNs, the question stands whether human-driver behaviour can also be estimated well-enough by those techniques, thus surpassing the current state of the art.
Objective and tasks The goal of this project is to evaluate and compare the descriptive potential of 1) an analytical statistical human driver model and 2) a black-box model realized using an artificial neural network. The student would be required to become familiar with the inputs and outputs of human driving behaviour models and the role they play in traffic simulations. Furthermore, the two models need to be calibrated using real life data and their performance needs to be compared. Completing the project would include the following steps:
1. Identification and preparation of the data set that will be used to create and test the models 2. Calibration of analytical statistical human driver model 3. Literature review to identify the most suitable neural network architecture and training of the model 4. Design of metrics and experiments to compare the two models 5. Results post-processing and analysis 6. Report writing
What we expect from you Object-oriented programming concepts C++ Statistical Modelling Scripting Basic understanding of discrete event simulations
What we offer you An international and multidisciplinary working environment Opportunity to work on a project with real-life relevance Work with researches from world-renowned Universities (TU Munich and NTU Singapore)
Enquires and How To Apply Please send your complete application including cover letter, CV, university transcripts and degree certificates to email@example.com