UROP Project

Vehicle Dynamics Modelling for Autonomous Racing with Deep Reinforcement Learning



Daniel Holder

Program Director UROP


+49 241 80-90695


Key Info

Basic Information

Project Offer-Number:
UROP International
Mechanical Engineering
Organisation unit:
Institute for Data Science in Mechanical Engineering
Language Skills:
fluent in english or german


In Deep Reinforcement Learning (DRL) an agent learns a control strategy from interacting with its environment. DRL has shown great success in controlling complex systems such as simulated robots or computer games. However, the application of DRL to real-world tasks such as vehicle control is a largely unanswered question that bears challenges beyond those of standard benchmark tasks. Our current research focuses on investigating and overcoming these challenges through the development of DRL methods tailored to the needs of real-world applications. We test those approaches on simulators of differing complexity as well as on hardware. Self-implemented simulators have proven to be an essential component of this testing pipeline as knowledge about the ground-truth dynamics is indispensable for verifying assumptions about the algorithms behavior.


Thus, we are looking for a UROP researcher that designs and implements a high-fidelity vehicle dynamics simulator that will be embedded in an autonomous racing framework for DRL algorithms. Starting from an existing simple kinematic bicycle model, the UROP researcher can investigate different modeling approaches to simulate vehicle dynamics. Based on the researcher's interest the correlation between simulator complexity and the learning success of different DRL agents can be investigated. The UROP student researcher will be asked to do the following, with the help of the supervisor: (1) Develop a brief overview of different modeling approaches for vehicle dynamics based on standard literature. (2) Implement one or more vehicle dynamics simulators in Python. (3) Integrate the simulator into the general autonomous racing framework. (4) Test the simulator with standard vehicle controllers and/or DRL approaches.


- Engineering and/or computer science background - Comfortable programming in a high-level programming language (e.g., Python) - Interest in dynamics modeling - Ready to problem-solve and work independently (with supervisor support)