In this project, we leverage the V2X network to mitigate and overcome the GNSS challenges. The main objectives are: 

• To develop data fusion methods to allow fusion of information from C-V2X, DSRC and other sensors including car radar, IMU, magnetic field measurements, cellular and partial GNSS signals (if any).
• To improve the performance of existing GNSS receiver by developing algorithms to augment GNSS with other in-vehicle sensors or V2X infrastructure and improve GNSS receiver processing algorithms for weak GNSS signal environment.
• Cooperation between multiple vehicles and RSUs via V2X information exchange to cooperatively perform vehicular localization. 
• To perform experiments on the V2X network test bed and develop a precise vehicular localization module or subsystem with industry stakeholders.

For more information, you may approach our Professor Tay Wee Peng.

Accurate localization with centimeter precision using data from multi-modal sensors, including camera images and lidar point clouds, is crucial for AV and challenging.  In this project, we use deep learning techniques to perform automatic feature extraction and location estimation. We generate a calibration map off-line by extracting landmarks from the vehicle’s field of view, while the measurements, from which the location is estimated, are collected similarly on the fly. 

The main objectives are: 

• To develop an end to end learned localization framework based on system-level machine learning techniques using multi-modal sensors.
• To show the usability and advantages of the end to end learned localization framework in typical autonomous driving use cases with comparison to traditional localization approaches.

For more information, you may approach our professor Tay Wee Peng.

The objectives of this project are:

• To develop a state space model for image and video information in the context of service provisioning in a V2X network.
• To develop machine learning methods for both inference and data privacy preservation in an ITS network, with application to the specific use-cases of protecting private information of drivers and passengers.
• To develop a privacy mechanism architecture that can be efficiently implemented in vehicles to achieve monitoring and service provisioning while ensuring inference and data privacy and low computational overhead.
  

Convolution neural networks (CNNs) have been used extensively in a large array of applications with notable success in image processing. There are attempts to extend the CNN architecture to general graphs. In this work, we develop various graph neural network approaches that mimic or subsume traditional CNN methods. For example, one approach is to decompose a graph into multiple parallel flows and perform convolutions on these flows. This approach subsumes both CNN as well as the popular graph convolutional network (GCN) architecture. 

For more information, you may approach our Professor Tay Wee Peng.

For more information, you may approach our Professor Wang Lipo.

Back to Research Areas