Sample Projects

Human Motion Tracking via Wearable Sensors and Machine Learning

Faculty Mentor: Prof. Brian K. Dean (PI),

The objective of this project is to develop a wearable sensor system capable of accurately tracking the motion of the human body. The system is to be designed such that it can be easily used by athletes while training or engaging in their sport. It should also be simple enough that the sensor brace, sleeve, sock, etc. can be taken home by patients doing home exercises for physical therapy. The students will be selecting sensors, designing signal conditioning/processing systems, developing methods for interfacing the sensors with the human body, and in later years designing an artificial intelligent system for processing information from multiple sensors to form an accurate 3D representation of the person in a computer simulation. Current work on this project includes the use of capacitive bend sensors, device level analog signal conditioning, an A/D converter, and an Arduino microcontroller for data collection and processing. Students will be able to choose which portion of the project to focus on. They can select alternative or additional sensors to test, improve the interfacing electronics, develop the Arduino code, or they can choose to work in multiple areas of the project.

GPU-Based Car Robot for Autonomous Driving Applications

Faculty Mentor: Shadi Alawneh (Co-PI)

Driving scenarios for self-driving cars are becoming increasingly complex with pedestrians, bicycles, road work, etc., so it is crucial that the models used for self-driving cars are able to respond to these scenarios as quickly as possible to ensure roads are as safe as possible. This project aims to build AI models using deep neural networks for autonomous driving applications, while using GPU computing to increase the speed of the training and inference processes of the AI models. The NVIDIA Jetson TX2 ‘Jet’ robot is used as the vehicle for autonomous driving, utilizing a camera mounted on the front to collect images in order to build the neural network. The output from the neural network was then sent to the Arduino in order to communicate to the robot which direction it needed to go to avoid any obstacles in its path.

The REU students will:

identify and collect the data that will be used for both training our deep learning model and testing the result,
analyze the data collected to select the best available dataset that will give us the best result to train our models,
develop the algorithms to train the deep neural network models with the available dataset,
write an algorithm to test the performance of the models, and
modify the hardware platform of the GPU-Based car robot to incorporate a Smart LiDAR Sensor which provides a more accurate measurements of the distance of the objects.

Socially Assistive Robots for Supporting Personalized K-12 Education

Faculty Mentor: Wing-Yue Geoffrey Louie

The world is rapidly ageing, and it is projected that there will be a decline in annual labor growth rate by twenty percent in the coming decade [5]. This labor decline has already led to shortages in the educational workforce. Such shortages will negatively impact the quality of education because either unqualified teachers need to fill these vacancies and/or teacher-student ratios will increase in the classroom. Consequently, there is an urgent need for investigating methodologies to help support the current as well as future demands on education services in the US. Socially assistive robots (SAR) are seen as a new unique strategic technology that will be important for addressing societies educational needs. Namely, SAR aim to provide assistance to human users through social interaction and not physical interaction. Current SARs are limited to the set of tasks that have been pre-programmed on the robot by an expert roboticist. Dr. Louie’s long-term research vision focuses on bridging the gap between non-expert end-users and robots so that robotic systems can be effectively utilized by the general population. Towards this vision, he aims to develop methodologies that enable K-12 teachers to easily teach or customize SARs to conduct novel tasks that support the educational goals of their classrooms.

In this REU project, students will:

design the user experience of the SAR teaching/customization system,
develop novel student-robot educational interactions for the classroom using the SAR teaching/customization system, and
conduct pilot trials of the developed educational interactions with students to investigate the educational efficacy of the interactions. This project will not only focus on the participation of undergraduates in robotics research but will also promote user-centered design of the SAR teaching system by having undergraduates unfamiliar with robots serve as users for pilot testing the systems.

In-vitro Measurements of Cerebrospinal Fluid Flow Rate for Hydrocephalus Treatment

Faculty Mentor: Hongwei Qu

One in every 500 newborns is affected by hydrocephalus – an incurable disease caused by excessive fluid that surrounds the brain. This accumulated fluid, known as cerebrospinal fluid (CSF), increases the intracranial pressure and enlarges the ventricles, which eventually lead to the damage of brain tissues and skull. Although it cannot be cured, hydrocephalus patients can be treated for a sustainable life. The most common way of treating hydrocephalus is implanting a surgical diversion of excess CSF using a synthetic tube called a shunt tube. In day-to-day neurosurgical practice, a common challenge is to achieve an optimal intracranial pressure through effective adjustment of the valve implanted with the shunt tube. The CSF flow rate is a critical parameter in setting the shunt valve for the brain pressure customized to each patient. The objective of this research project is to investigate a non-invasive mechanism for CSF flow rate measurement, which may potentially lead to a convenient and reliable solution in hydrocephalus treatment. Thermal time of flight (TTOF) approach is proposed to accurately measure the flow rate in-vitro. The measurement system consists of a thermal excitation source and highly sensitive temperature sensors placed at the measurement locations downstream along the modified CSF tube. The modified tube will feature reduced thermal resistance at the measurement spots for sensitivity enhancement without altering the CSF flow dynamics. To mitigate possible thermal contamination caused by other excitation sources such as an ice pad, a solid-state thermoelectrical cooling (TEC) device with reduced contact area is energized as the thermal excitation. Figure 3 shows a schematic of the measurement system, in which the TEC, the CSF flow, thermal path for temperature measurement at spots T1 and T2 are illustrated. The measurement will be conducted with artificial skin for system accuracy.

The participating students will conduct research in the following aspects:

using finite-element analysis (FEA) to examine the CSF flow and temperature profiles under a thermal excitation,
design and implementation of a TEC module for quick and stable temperature excitation at 2 °C which is safe for human skin,
performing benchtop tests of the system and TTOF measurement with representative liquid, and extrapolation of the CSF flow rate.

Plane Wave Ultrasound Image Formation with Deep Learning

Faculty Mentor: Alycen Wiacek

Plane wave ultrasound imaging has revolutionized the ability to measure fast biological functions such as the fetal heartbeat, non-invasively, using ultrasound. However, to achieve a high frame rate, only single plane waves are used to create images, resulting in poor image quality and making diagnostic decisions difficult. By using multiple plane wave acquisitions, the image quality can be improved at the expense of frame rate. To improve upon this trade-off between image quality and frame rate, recent research has focused on the development of deep neural networks to input single plane wave images and output images with image quality comparable to those generated with multiple plane waves. The objective of this project is to develop a deep neural network that is capable of generating high quality images using only a single plane wave. The main goals of the proposed project include: (i) Design a neural network to learn optimal image reconstruction with input a single plane wave image and output an image of higher quality, (ii) develop realistic and physics-based data augmentation methods to aid in training, (iii) quantitatively compare network-generated images to ground truth images generated with multiple plane waves and to state-of-the-art neural network models using well-known image quality metrics (e.g., contrast, SNR, CNR, gCNR), (iv) measure network complexity and frame rate and compare to state-of-the-art image reconstruction algorithms.

Students working on this project will actively participate in the following research tasks:

explore existing deep neural network models and select appropriate model for the imaging task,
develop a pipeline for data preprocessing using the CUBDL database,
develop data augmentation methods to aid in network training,
train and test the network on simulated, phantom, and in vivo data,
compare network-generated images to state-of-the-art algorithms in terms of image quality and frame rate

Page updated

Report abuse