Automated Vehicle Simulation and Digital Twin
Mentor: Brendan Morris
Modern AV design and testing relies heavily on vehicle simulation to create realistic environments scenarios for repeatability or for safety. Once algorithms have been perfected in simulation they can then be implemented in a real-vehicle with confidence. This project aims to implement a driving simulator on our driving simulator motion platforms. The goal is to connect three different driving simulators (two with three degrees of motion and a desktop-based) for cooperative simulation and testing. The simulation environments under consideration is Carla driving simulator and Autoware for the autonomy software stack.
Required skills: Programming (preferably Python and C++), preferred skills: ROS and driving simulation (e.g. from gaming)
Areas: Autonomous Vehicles, Digital Twin, 3D Modeling
Automated Translation Calibration for Autonomous Vehicles
Mentor: Chuck Tessler and Prashant Modekurthy
This project asks students to develop an automated calibration system for cameras mounted to autonomous vehicles. The platform is a 1/10th scale vehicle running the popular Robotic Operating System (ROS) as part of F1Tenth racing.
Required skills: Familiarity with python (or C) and a limited background in linear algebra
Areas: Computer Vision, Real-time computation
Real-Time Transmission Line Inspection based on Deep Learning Methods
Mentor: Mei Yang
Ensuring effective power grid monitoring and inspection is crucial for preventing power failures and potential blackouts. As the power grid continues to expand, finding solutions for providing automatic, accurate, and real-time inspection of transmission lines becomes a critical challenge at both regional and national levels. Unmanned aerial systems (UAS) have emerged as a valuable tool for regular transmission line inspections, offering advantages such as high flexibility, cost-effectiveness, and the ability to cover vast areas during both day and night. UAS-based transmission line (TL) inspection systems have recently emerged, incorporating various visual remote sensing technologies and deep learning (DL)-based computer vision techniques. However, due to insufficient annotated image datasets and limited communication bandwidth & computation power, real-time detection remains a big challenge. In this project, we propose the development of a DL-based UAS framework for real-time inspection of TL. The objectives of this project are: 1) Identify high-risk areas for TL failures and construct a localized dataset of TL components specific to the target region. 2) Develop a real-time inspection system of TL failures using transfer learning based on a general dataset of TL components and transfer the trained model to the localized dataset for improved performance. And 3) Design a lightweight CNN model with reduced complexity and implement it on a Raspberry Pi platform to enable efficient and practical deployment on UAS. The proposed framework and DL-based approaches will help automate the inspection of transmission lines and improve the timeliness of component failure detection.
Required skills: Python
Areas: Machine Learning, Internet of Things