Date of Award
August 2021
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Mechanical Engineering
Committee Member
Yue Wang
Committee Member
Yiqiang Han
Committee Member
Mohammad Naghnaeian
Abstract
For the past 30 years, autonomous driving has witnessed a tremendous improvements thanks to the surge of computing power. Not only did we witness the autonomous vehicle navigate itself safely in the urban area, stories about more diverse autonomous driving applications, such as off-road rally-style navigation, are also commonly mentioned. Just until recently, the exponential increase in GPU and high-performance computing technology has motivated the research on autonomous driving under extreme situations such as autonomous racing or drifting.[25] The motivation for this thesis is to offer a brief overview about the main challenge of autonomous driving control and planning in racing scenario along with the potential solutions.
The first contribution is using koopmam operator and deep neural network to perform data-driven system identification. We then design optimal model-based control which is based on the learned dynamics alone. Based on our new system identification algorithm, we can approximate an accurate, explainable, and linearized system representation in a high-dimensional latent space, without any prior knowledge of the system. In this case, the learned vehicle dynamic automatically involves the information that is normally difficult to obtain, including cornering stiffness, tire slip, transmission parameters, etc. Our result shows that our koopman data-driven optimal control approach is able to deliver better tracking accuracy at high speed compared to the state-of-art vehicle controllers.
The second contribution is an iterative learning and sampling algorithm that can perform minimum-time optimization of the global racing trajectory(aka racing line) within the limit of tire friction. This trajectory optimization algorithm is not only proven to be computationally efficient, but also safe enough for the onboard RC vehicle’s test.
The research achievements we made for the last two years not only enables the F1TENTH racing team of Clemson University Mechanical Engineering Department to finish top 5 in both virtual autonomous racing hosted by IFAC and IROS congress, but also offer us the opportunity to join ICRA 2021 Autonomous racing workshop to present our work and being awarded the joint best paper. More importantly, these contributions proved to be functional and effective in the on-board testing of the real F1TENTH robot’s autonomous navigation in the Flour Danial basement. Finally, this thesis will also include discussions of the potential research directions that can help improve the our current method so that it can better contribute to the autonomous driving industry.
Recommended Citation
Wang, Rongyao, "Data-driven System Identification and Optimal Control Framework for Grand-Prix Style Autonomous Racing" (2021). All Theses. 3601.
https://open.clemson.edu/all_theses/3601