Fung Institute End of Year Showcase 2024

Team: Rishita Khandwala [NE], Nishita Phillip [MSE], Ibrahim Albrahim [MSE], Matthew Johansson [NE]
Advisor(s): Peter Hosemann
Project ID: 146

In an effort to promote the widespread adoption of sustainable energy, we aim to encourage the mass commercialization of nuclear power plants. However, the most common materials used in nuclear reactors, zirconium alloys, are expensive amongst other drawbacks. High Entropy Alloys (HEAs) are a promising alternative, exhibiting impressive properties. Using an arc melter, we are developing an alloy tailored for these extreme environments. Our HEA will provide alternatives for existing materials and extend applicability to extreme environments even beyond nuclear engineering applications.

Team: Adam Chiu [ME], Chang Li [ME], Xitong Wei [ME]
Advisor(s):Sara McMains
Project ID: 110

We are developing a machine learning algorithm surrogate for the time-consuming computer simulation that is traditionally used to predict the quality of 3D printed metal parts in order to save time and materials by mitigating additive manufacturing failures. We aim to predict the unprintability score to an accuracy within 10% error of Finite Element Method (FEM) simulation results to replace FEM in the manufacturing process, enabling faster delivery to the customer. We will be using Ansys Mechanical and Abaqus to obtain FEM results from a diverse range of parts as training and validation data for our PyTorch CNN.

Team: Chongyuan Li [ME], Arjun Rajasekaran [MSE], Cue Li [ME], Mu He [ME], Lily Chang [MSE]
Advisor(s): Rayne Zhang, Victor Couedel, Haotian Lu
Project ID: 82

Vascular thrombosis is a condition that results in blocked blood vessels. Stents, deployed through minimally invasive procedures, play a crucial role in preventing vessel closure and reducing clot formation. 3D printing via stereolithography enhances stent design, offering precision, mass production efficiency, biocompatibility, and customization. We innovatively incorporated ultrasound capability into 3D-printed polymer stents, enabling real-time monitoring of blood flow. These stents, designed with shape memory and piezoelectric properties, aim to generate ultrasonic power for post-implantation surveillance. This personalized, efficient, and biocompatible approach addresses the complex challenges of vascular thrombosis, advancing medical technology and improving patient outcomes.

Team: Chenyi Hu [MSE], Jainabh Hasanali Kerosenewala [MSE], Teresa Rodriguez Sanchez [ME], Henry Huang [ME]
Advisor(s): Tarek Zohdi
Project ID: 114

Conventional materials have limitations in terms of shock absorption and viscous energy dissipation, leading to in reduced lifetime of materials and safety concerns especially in the automotive/aerospace industry. Meta-materials, artificially engineered to have tailored properties, can improve shock energy absorption at impact up to 200% better than the status quo. Our goal is to design a unit cell of the meta-material contained in an optimal geometry to improve shock absorption using the simulation software LSDyna and optimizing the most relevant design parameters via machine learning techniques in Python.

Team: Shao Che Chen [MSE], Enrico Hariono [MSE], Enrico Milletti [MSE], Kirthi Rachakonda [MSE], Nandni Sinha [MSE], Alanna Smith [MSE]
Advisor(s): Ethan Escowitz
Project ID: 133

Currently around 17 million tons of textiles are produced each year, 11 million of which are sent to the landfill. To lower the textile industry’s carbon footprint, Artefact aims to manufacture composites made from upcycled textiles and various other natural and recycled materials for consumer product applications. To advance this mission, our team is developing an end-to-end manufacturing system by conducting materials research, optimizing composite manufacturing processes, and developing testing and validation methods. The current goal of the project is to produce bio-composite eyewear frames that exhibit competitive strength, durability, and aesthetics to synthetic materials currently used in the market, with an eye towards transferring the core technology to other industries and applications in the near future.

Team: Mateo Navarro Goldaraz [MSE], François Logak [NE], Maya Roma [MSE]
Advisor(s): Peter Hosemann, Alexandros Spyromilios
Project ID: 150

Diamond is a unique material with many extreme properties useful for semiconductors and heat dissipation in power electronics. Diamond is the hardest known material, making it very difficult to cut or polish, which is why our team is scanning parameters for processing lab grown diamonds with lasers to improve the speed, effectiveness, and cost. Our project expands on previous research regarding laser environment, angle of incidence, and energy fluence using a novel femtosecond laser process, new environments, and new characterization techniques. Our results will impact the growing synthetic diamond industry, propelling diamond based technology accessibility.

Team: Tse-Yu Chang [ME], Wenlan Fan [ME], Yifan Liang [MSE], Shuinan Liu [ME]
Advisor(s): Peter Hosemann, Matthew P. Sherburne
Project ID: 145

Inspecting the metallic coating thickness in car brake disc manufacturing is essential for quality control and product certification. The traditional inspection process requires a labor-intensive and time-consuming step of cutting the disc for proper sample preparation, raising overall production costs while significantly delaying feedback to process control. Our team will develop a novel non-destructive technique using high-energy optical sources to allow for rapid characterization of the coating thickness. The successful implementation of this new technique will reduce manufacturing costs and improve efficiency.

Team: Chinmay Mahesh Bukinkere [ME], Olivia Chen [MSE], Yifan Duan [MSE], Aaron Tang [ME], Prajwal Vempatti [MSE]
Advisor(s): Marco Maurizi, Xiaoyu (Rayne) Zheng
Project ID: 88

Young athletes are required to wear protective gear when playing sports. For example, lacrosse, where high-speed projectile-induced chest trauma, also known as Commotio cordis, is a rare but deadly occurrence. Currently NOCSAE-certified bulky protective foam vests are commercially available. Using machine learning and 3D designing in CAD software, new designs of novel metamaterial geometries have been generated. The metamaterials are currently being 3D printed, tested, and evaluated for high impact resistance, and are predicted to have superior impact resistance compared to foams that are currently utilized in protective sports equipment.

Team:Sydney Chen [CEE], Caoying Huang [CEE], Thomas Perera [IEOR], Mike Santos [CEE]
Advisor(s):Jasenka Rakas
Project ID: 156

Urban Air Mobility (UAM) is a new aviation concept that will transport people and goods in urban metropolitan areas using electric Vertical Takeoff and Landing aircraft (eVTOL).
The purpose of this project is to develop a framework/tool for vertiport site and vertiport size selection to support the implementation of eVTOL services in the airport access, able to balance passenger experience criteria and UAM operational constraints.Estimating passenger demand for eVTOL operations between vertiports and airports, and determining the maximum number of houry eVTOL operations at vertiports are important metrics for safe and reliable UAM operations.

Team: Ellie Ko [ME], Demin Li [ME], Yingze Hou [EECS], Run A [EECS]
Advisor(s): Samuel Gonzalez
Project ID: 61

Mining sites are existing operational areas that can extract valuable natural ores and minerals on earth from various geographical locations. Being able to extract their resource types as well as gauging potential hazardous conditions in surroundings can be essential for investment decisions. Our goal is to provide a machine learning model that can identify geographical metadata to help investors gain more geo-related knowledge, thus avoiding potential investment risk. Technically, we are utilizing U-Net to segment the contour of mining sites and Yolo to classify the resource type, along with other helper Python packages to realize the identification of these metadata.

Team:Yaqiao Jiang [IEOR], Lingxi Kuang [IEOR], Jeevesh Guha Natarajan [IEOR], Haohan Wang [IEOR], Huanhui Ye [IEOR], Zeqi Zhang [IEOR]
Advisor(s): Lizeng Zhang
Project ID: 172

Our Capstone Project focuses on using a Convolutional Neural Network (CNN), trained using images of established trading patterns like “Head & Shoulders” and “Inverse Head & Shoulders”, to enhance the trading capabilities of Trading Desks. We also leverage Generative Adversarial Networks (GANs) to effectively enhance the classification potency of the CNN, by delivering a more comprehensive representation of diverse market patterns.

Team: Ryan De Koninck [IEOR], Sining Huang [EECS], Zixuan Huangfu [IEOR]
Advisor(s): Karl Sahyoun
Project ID: 31

Traditionally, sourcing investment opportunities for venture capital and growth equity firms has been laborious due to limited access to private company data. With the emergence of databases like Crunchbase, housing such data, leveraging machine learning presents an opportunity to streamline this process. Our project centers on developing a machine learning model, trained on a collection of datasets, to forecast a company’s likelihood of securing funding within a specified timeframe. By bypassing much of the labor-intensive sourcing process, we aim to boost the efficiency of Bregal Milestone’s investment team, enabling swift, data-driven decisions and a competitive advantage.

Team: Alexandre Soppelsa-Metzelard [IEOR], Anushka Baid [IEOR], Paul Faverjon [IEOR], Capucine Hustin [IEOR]
Advisor(s): Lizeng Zhang
Project ID: 170

In today’s dynamic financial landscape, characterized by volatility and uncertainty, accurate prediction of stock returns is essential. Traditional models often struggle to capture complex market dynamics, prompting us to explore Deep Learning techniques like LSTM, GRU, and Transformers. By benchmarking against traditional models, we aim to improve stock price prediction accuracy, empowering stakeholders with advanced tools for investment strategies and risk management. We aim to bridge the gap between the demands of modern markets and the capabilities of existing models, helping one make smarter investment decisions and maximize profits.

Team: Yang Gao [ME], Yixiao Kang [EECS], Daniel Barron [EECS]
Advisor(s): N. Benjamin Erichson, Michael Mahoney
Project ID: 160

Stable Science is dedicated to advancing the precision of scientific imaging by leveraging AI-powered diffusion techniques to enhance image resolution. Our approach meticulously calibrates low-resolution inputs to generate detailed outputs that strictly conform to the governing physical laws of the depicted phenomena. Our custom model ensures that each enhanced image is not only clearer but also a true representation of its real world counterpart. Instead of using traditional neural networks methods, we applied generative AI models. We is poised to deliver a tool that expedites scientific analysis, supporting researchers in achieving accurate, reliable results swiftly and sustainably.

Team: Lakshya Aggarwal [EECS], Jane Mathew [BIOE], Yi Wang [IEOR], Gloria Wang [IEOR], Chan Hyuk Yang [IEOR
Advisor(s): Lucas Allen, Adheesh Shenoy, Mitchell Ding, Tarek Zohdi
Project ID: 122

Over 145 trillion dollars worldwide are under management, with sovereign wealth funds being a subset invested by countries with the aim of having returns that can fund pensions, infrastructure, healthcare, and all manner of services provided by the state. Our project intersects finance and NLP, harnessing cutting-edge LLMs like Mistral and Mixtral to extract insights from financial reports. We’ve curated datasets and trained models to discern from reports critical details such as investors, investment amounts, and dates. By facilitating expedited analysis of fund reports with our LLMs, we hope to aid in the well-management of these sovereign wealth funds.

Team: Sameer Bhuvaji [ME], Yinu Guo [ME], Navin Vijey Raj Jeyaselvan [ME], Jeffrey Tsang [ME]
Advisor(s): Francesco Borrelli, Alessandro Pinto
Project ID: 141

Current limitations in data pose a significant challenge for NASA’s Endurance lunar rover mission. These limitations can lead to inconsistent mission planning, resulting in cost overruns. Team Endurance’s project provides a robust approximation of the mission time and plays a crucial role in optimizing performance and mitigating potential risks. We have built a functional simulation in C++, and have successfully implemented the mobility and communications subsystems to calculate the distribution of mission completion time. Our data will play an impactful role in evaluating the rover’s performance, thus increasing the reliability of the Endurance mission.

Team: May Tang [EECS], Stefan Shi [EECS], Jimmy Gan [EECS]
Advisor(s): Zheng Liang
Project ID: 91

In this project, our aim is to thoroughly investigate the efficient deployment of large language models (LLMs) concerning execution time, memory footprint, and computational resources. Students explore various stages, including fine-tuning/training, and inference, to optimize their performance.

Team: Jhan-Shuo Liu [EECS], Nozomu Kitamura [CEE], Qingyang Hu [EECS]
Advisor(s): JD Margulici
Project ID: 24

Our project aims to revolutionize urban planning decision-making by introducing a low-code framework that simplifies the creation of data-driven tools. Specifically, we target two critical challenges as proof of concept: mitigating traffic congestion through predictive traffic flow analysis and streaming data transformations for decision-making. Utilizing a low-code approach with digital twins, we aim to construct an accurate machine-learning model and create an intuitive dashboard for data analysis with cloud computing. This endeavor seeks to alleviate traffic bottlenecks and demonstrate the potential of low-code platforms for non-technical users to develop advanced applications, fostering broader innovation in urban planning.

Team: Apratim Banerjee [EECS], Yuan Zhang [ME], Tianqi Zeng [ME], Shiyun Huang [ME]
Advisor(s): Bike Zhang
Project ID: 123

While Large Language Models (LLMs) show promise in reasoning and action planning in robotics, their effectiveness diminishes in describing real-world scenarios and complex instructions. Current LLM-based task planning often lacks critical analysis and optimization, impacting efficiency in industries such as logistics, manufacturing, and distribution.
Our work involves leveraging LLMs like GPT-4 to decompose action tasks from natural language input and feed it via our native coding environment for the robot to execute. In addition, our novel framework, the Large Language Multi-tasking Optimizer (LLMAO) is capable of multi-tasking and long horizon planning, by modeling sequential dependencies between actions and mapping the relationships onto a graph optimization framework. Our model is proven to be more accurate and efficient to previous LLM path planning models.

Team: Yunlu Li [IEOR], Ken Akers [EECS], Pranav Agarwal [IEOR], Anand-Arnaud Pajaniradjane [IEOR]
Advisor(s): Clement Ruin
Project ID: 106

Nearly 30% of online retailers are using AI-powered chatbots for quick and cost effective customer service solutions, but strong dependence on slow, expensive human feedback impedes their development cycle. The bot2bot team aims to accelerate this process by building large language models (LLMs) that simulate realistic customer conversations, thus empowering companies to efficiently develop chatbots that adapt to ever-changing customer needs, business strategies, and market conditions. Our framework pairs state-of-the-art retrieval augmented generation (RAG) and fine-tuning techniques with open-sourced LLMs to tailor conversation semantics to each company.

Team: Omair Alam [EECS], Alice Lee [EECS], William Almy [EECS], Hao-Ming Hsu [EECS]
Advisor(s): Anant Sahai
Project ID: 85

In today’s interconnected world, radio spectrum signals surround us, yet there exist noticeable limitations in the data systems created to access, monitor, perform AI experiments, and contribute to this analog data. To democratize the access and usage of spectrum data, we have built SpecPipe, a distributed AI/ML data pipeline. This platform’s core values of accessibility, extensibility and scalability ensure that individual users can start to work with radio data with inexpensive hardware, minimal configuration, and a smooth onboarding process. We have accomplished this goal of improving access to spectrum data by building SpecPipe as an open-source project free for people to access and use, with easy to follow documentation, and a plethora of startup examples that allow users to understand our framework interactively.

Team: Xiaolei Ye [ME], Jui-Che Chang [ME], XiuYu Zhang[EECS], Yue Fang [ME]
Advisor(s): Kosa Goucher-Lambert
Project ID: 151

3D objects have wide applications across different industries, such as in design, manufacture, games, and AR/VR. Despite the growing interest in 3D modeling, designing 3D objects from scratch takes time and effort.To allow 3D designers to access a wide range of 3D objects for creativity, we propose a ML assisted framework, i.e., CLAS (Capture, Label, Associate, and Search), to enable fully automatic retrieval of 3D objects based on user specifications. CLAS provides an effective and efficient method for any person or organization to benefit from their existing but not utilized 3D datasets.

Team: Florence Rusly [IEOR], Louis Paris [IEOR], Bruno Tabet [IEOR]
Advisor(s): Lee Fleming
Project ID: 175

Being able to predict scientific successes of scientists can help hiring committees, as well as government funding agencies, and venture capital firms in investment decisions. We are trying to predict the future yearly citation counts and breakthroughs (papers that are cited more than 100 times 3 years after publications) of 1000 scientists in the U.S. working in the Biotech industry. Using past publications, patents, citations, we trainrf various machine learning models (ensemble methods, neural networks, etc.) to make the most accurate predictions. Our best performing models can predict which scientists will have a breakthrough with a 0.85 accuracy.

Team: Yuanbo Chen [EECS], Chengyu Zhang [EECS], Jason Wang [EECS], Xuefan Gao [IEOR]
Advisor(s): Avideh Zakhor
Project ID: 13

3D indoor scene reconstruction is a task with crucial applications in preserving historical sites and recreating AR/VR scenes of the real world. Recently, drones have been utilized in reconstruction due to their ability to explore areas that are hard to reach for other robots. To leverage these benefits, we utilize a drone and 360 camera to reconstruct a 3D virtual environment that is quickly viewable from any computer. Our novel pipeline utilizes a divide-and-conquer approach that scales efficiently to large scenes. Further, we employ gaussian splatting to achieve higher quality results and lower training time than previous methods.

Team: Alexandria Lin [EECS], Geneivie Nguyen [ME], Manshi Yang [EECS], Yucheng Huang [EECS]
Advisor(s): Tiffany Tao
Project ID: 74

Furiends, an augmented reality (AR) mobile game created by health-tech company, Blue Goji, utilizes AR to promote physical activity through a furry friend. According to studies, it is challenging to keep users active and engaged while exercising. To tackle this problem, Furiends engages users by offering a unique pet-raising experience and transforms step counts into in-game currency, enhancing user interaction with the game. Our plan is to launch the game on Apple iOS by migrating from its previous server and updating the platform version and AR scripts.

Team: Yukun Song [EECS], Kathy Zhuang [EECS], Rui Li [EECS], Zoe Zhou [EECS]
Advisor(s): Allen Yang
Project ID: 16

Our project aims to assist astronauts’ operations of robots in space by visually recognizing the location and orientation of the robot. With our algorithm, we are able to append augmented reality (AR) interfaces beside the robot in the astronauts’ view with using HoloLens. We leverage a motion tracking system and a high-quality camera with depth sensors to collect datasets and train the transformer-based neural network. Specifically, the solution provided by our project enables a faster and more intuitive interaction method between astronauts and the Leo Rover, which can also be extended into daily use cases of human-robot interactions.

Team: Angel Plaza Carreras [ME], Benjamin James Li [NE], Claire Elizabeth Garlington [MSE], Jack Fiske Kirkwood [NE]
Advisor(s): Max Monange, Massimiliano Fratoni
Project ID: 15

Today’s carbon and energy challenges necessitate a new source of clean energy that is cheap, reliable, and safe. Nuclear fusion shows renewed promise after recent successes at the National Ignition Facility. Viable reactors must replenish the tritium fuel they use by capturing neutrons in a “blanket” to “breed” new tritium. We are a capstone team at UC Berkeley working with the laser fusion startup EX-Fusion to design and simulate their demonstration reactor blanket. Our team is pioneering new technology for a self cooled lithium-lead blanket with novel structural materials via iterative design and simulation, as shown on the left. This work serves as the foundation for EX-Fusion’s 200 megawatt electricity-producing demonstration reactor in 2035.

Team: Felix Ellwood (ME), Magnus Frankevoort (ME), Antonia Ginsberg-Klemmt (ME), Qixi Liao (BIOE), Sulav Parajuli (ME)
Advisor(s): Tarek Zohdi
Project ID: 161

Over 90 million American households, representing 77 percent of the population, are unable to access solar energy due to factors such as renting homes, policy limitations, and physical barriers, highlighting the urgent need for broader accessibility. Our goal is to optimize the design of the commercially available mobile solar carport appliance, the MEGA, and create a detailed assembly instruction manual with technical drawings for the various components for industry production. Using SolidWorks, we are generating CAD prototypes, modifying geometric parameters, decreasing component numbers, and reducing overall cost, all while maintaining the structural integrity to make PV accessible to all.

Team: Christopher Simotas [ME], Casey Leong [ME], Dingxiang Peng [MSE], Davide Franceschini [NE], Georgios Chantzakos [ME]
Advisor(s): Reza Alam
Project ID: 89

To meet the global carbon neutral goal, ocean wave power emerges as a pivotal player due to its vast and consistent source of energy. However, its complexity and harsh nature have driven the industry towards expensive, large, and material-intensive designs, which have limited its adoption. Our team aims to build a prototype wave energy point absorber that leverages inflatables and a streamlined power take-off system to drastically reduce cost. The final result is a cost-effective, compact, and easily deployable product that can produce affordable energy.

Team: Shofi Latifah Nuha Anfaresi [CEE], Daniel Garcia Mijares [ME], Prin Seetapan [ME], Guanyu Zhao [ME]
Advisor(s): Van Carey
Project ID: 78

By the year 2050, agricultural water scarcity is projected to rise in over 80% of the world’s agricultural land. This exacerbates the difficulties faced by more than 122 million individuals globally struggling with food insufficiency. Airborne Snow Observatories Inc. has developed snow maps capable of predicting water availability from snow. However, errors in the data observations necessitate manual correction by highly specialized engineers. Our team is developing an automated algorithm to compress this knowledge, enabling the identification and elimination of noise. This algorithm will reduce time and resource usage, ultimately enhancing decision-making for data users.

Team: Mason Rodriguez Rand [ME], Yashvardhan Saravan Nalina [ME], Andres Arraiz [MSE], Crystal Yang [MSE], Ben Sanders [MSE]
Advisor(s): Simo Mäkiharju
Project ID: 65

Direct Air Capture (DAC) is a process that extracts CO2 from ambient air by reacting it with liquid chemistry. This reaction occurs on the surface of the packing material. Currently, DAC costs 3 times more than what is commercially viable. A primary reason for this is that existing packing materials are not tailored for DAC technology. Therefore, we are using iterative experimental design to evaluate current packing materials and relate properties like shape, porosity and contact area to the CO2 capture rate. Using our key takeaways, our team will design an efficient and scalable packing material that maximizes capture rate.

Team: Max Kasteel [CEE], Arsh Anis Khan [ME], Emanuel Pytlik [EECS]
Advisor(s): Reza Alam
Project ID: 90

Our primary objective is to significantly reduce fuel consumption in the shipping industry. Achieving this goal involves deploying autonomous ocean drones equipped with advanced sensors. These drones autonomously navigate, optimizing their positioning to precisely measure key wave characteristics, including height, frequency, and direction. By integrating this data with cutting-edge routing algorithms developed at UC Berkeley labs, we aim to achieve a substantial reduction of over 20% in shipping fuel consumption. Combining cutting-edge technology and scientific research, this project seeks to mitigate the environmental impact of the shipping industry and contributes to the broader effort to combat global climate change.

Team: Jasper Chua [ME], Jeffrey Dible [ME], Sofia Lopez [ME], Jorge Pacheco [ME]
Advisor(s): Anthony Joseph
Project ID: 10

Flying a helicopter running on contaminated fuel can cause catastrophic damage. Thus, it is part of the Army’s pre-flight checklist to take and analyze a fuel sample before any flight. Currently, soldiers do this using a primitive system of siphoning fuel from the tank via thumb pump (left), and depositing the fuel into a hand-held mason jar. This process is slow, difficult to perform, and leaves soldiers with fuel on their skin or uniforms. Our team aims to design, prototype, and test an enhanced fuel sampling system (right) that is faster, more ergonomic, more reliable, and safer than the current solution.

Team: Ivan Bondarenko [ME], David Kwok [ME], Leon Marre [IEOR], Sahil Mehta [IEOR]
Advisor(s): Lee Fleming
Project ID: 173

There are a lot of different companies building and operating EV charging stations and even more companies that offer payment solutions via cards or apps with different pricing models. Not every card enables access to every charging station as this depends on contracts between the different companies. This situation leads to confusion and inconvenience for all EV drivers because they need to carry multiple cards to ensure access to as many charging stations as possible and can’t keep track of the most economical way to charge. Our capstone aims to create a startup which enhances and simplifies the EV charging payment experience with an all in one mobile app. Our key feature is the consolidation of different charging cards into one application which then uses an algorithm to find the cheapest payment method for charging at a given station.

Team: Shivam Gupta [ME], Riddhi Sera [BioE], Sofia Haile [BioE], Yi Hong Liu [BioE], Markus Bauer [EECS]
Advisor(s): Shawn Shadden
Project ID: 135

Medical imaging has revolutionized cardiovascular medicine. However, medical imaging mainly provides anatomical information, whereas diagnosis (and treatment planning) of cardiovascular problems requires functional information. This project leverages machine learning to automate the development of patient-specific computer models from medical image data to support the simulation and analysis of functional information.

Team: Bo-Yen Chang [BIOE], Sophie Furlow [BIOE], Florian Kristof [BIOE]
Advisor(s): Edilberto Amorim De Cerqueira Filho
Project ID: 19

The standard of care in ICUs is to observe comatose patients for several days before making treatment decisions, but early action can lead to better recovery outcomes. We are deploying a model trained on historical patients’ neurophysiological data and doctor-written notes to predict the likely recovery outcome of new patients. The prediction is a decision support tool that helps physicians make earlier treatment decisions to help patients reach a higher level of functional independence post-coma. Using Python, we cluster the similarity of brainwave patterns and use natural language processing to correlate keywords in patients’ notes to their likely recovery outcomes.

Team: Angana Dasgupta [BioE], Natalie Saadeh [BioE], Boyan Yin [BioE], Isabella Lopez [BioE]
Advisor(s): Syed Hossainy, Kevin Healy
Project ID: 116

In the United States, someone has a heart attack approximately every 40 seconds, presenting a profound public health concern (CDC). We are pioneering a heart-motion stimulated piezoelectric patch to revolutionize cardiac rehabilitation. When strategically placed on infarcted cardiomyocytes, this patch harnesses the heart’s natural mechanical movements to induce a small voltage that regenerates damaged cells. Leveraging the unique properties of PVDF films, this innovation stimulates heart cell regeneration for survivors of heart attacks. By integrating piezoelectric materials with cardiac biology, our project holds promise for millions, marking a breakthrough in advancing recovery, improving life quality, and revolutionizing cardiac health.

Team: Qianyun Lin [BIOE], Zehao Wang [BIOE], Chenjie Ye [BIOE]
Advisor(s): Anthony Joseph, Dave Fabry, Majd Srour
Project ID: 111

According to WHO, over 1 billion people are at risk of permanent, avoidable hearing loss due to unsafe listening practices. A system that proactively warns users of overexposure of noise can lead them to a healthier lifestyle and prevent further deterioration of hearing. We would like to create an intellectual health care manager using Generative AI that can not only prevent further hearing loss, but also improve social engagement of our users.

Team: Anna Luszczak [BIOE], Daya Rao [BIOE], Tyler Hodge [BIOE], Jamie Jiang [BIOE].
Advisor(s): Syed Hossainy, Mohammad Mofrad
Project ID: 115

Mitral regurgitation (MR) is the leakage of blood back into the left atrium during systole that affects over 6 million people within the US and is a global health concern1. Our project aims to computationally analyze the strain effects on the left ventricle combining the finite element software ANSYS with fluid mechanics software CircAdapt, enhancing left ventricle biomechanical response understanding, and facilitating treatment assessment and improvements. Our project objective also includes an overarching experimental model that can validate both literature data as well as computational findings. This ensures our models for mitral valve regurgitation are accurate, robust, clinically relevant, and innovative.

Team: Jessica Boetticher [BIOE], Adam Duong [BIOE], Ben Margolis [ME], Panos Pardalidis [BIOE]
Advisor(s): Hannah Stuart
Project ID: 54

Patients with C5-C7 spinal cord injuries have limited grip strength, leading to difficulty performing routine tasks. Current solutions involve costly technologies and complex mechanisms to enable patients to engage in activities of daily living. We are enhancing patients’ independence through a low-cost assistive grasping device driven by electric power that does not impede natural grasping function. Our grasping device consists of two supernumerary fingers located on the back of the wrist and utilizes flexor sensing to detect wrist bending, which closes the fingers for grasping objects.

Team: Clemence Rausa [BIOE], Amanda Anil [BIOE], Harshvardhan Dhanpal Ankalkhope [ME], Noor Gulrajani [BIOE]
Advisor(s): Stephen Hill, Tarek Zohdi
Project ID: 2

The ACL is a primary stabilizer of the knee, preventing excessive extension and rotational movement. Female athletes are disproportionately impacted by non-contact ACL injuries, at least 2 times more likely to experience a tear than males. Understanding this injury requires a comprehensive, noninvasive, and accurate analysis of female ACL dynamics. Using motion capture, we track markers placed at precise anatomical landmarks of the knee during various maneuvers. In deriving ACL strain throughout these movements, high-risk activities can be identified. With 200,000 annual ACL reconstructions in the United States, these findings bring awareness to athletes and injury prevention methods.

Team: Nguyen Kha Ai Vo [BIOE], Serene Azafrani [ME], Leslie Pu [EECS], Hugo Hakem [BIOE]
Advisor(s): Ryan Kaveh, Elise Scipioni
Project ID: 62

The measurement and recording of the electrical activity of the brain through the use of Electroencephalography (EEG) has the promise of personalizing rehabilitation treatment programs. However, traditional EEG setups are complicated and, as a result, limited to lab settings. In response, MZR Neurotech has developed EEG wearables in the form of a Headband and an Earbud. Our Team utilizes these EEG wearables in partnership with Blue Goji, a startup focused on cognitive-motor rehabilitation, to confirm the feasibility of continuous EEG data collection in motion environments and ultimately add brain alpha wave recording as a metric for tracking a patient’s cognitive health state.

Team: Chih-Yao Chan [ME], Megan Dao [BIOE], Quentin Daurat [ME], Andy Kim [ME], Tadakatsu Nakashita [ME]
Advisor(s): Victor Detavernier, Bryan Tran, Gabriel Gomes
Project ID: 32

The project revolves around an omnidirectional treadmill, which enables users to walk in multiple directions. The current phase of the project builds upon the work of previous capstone teams who contributed to the prototype’s development. The objective is to leverage the expertise and fresh ideas of this year’s team to further enhance the treadmill. The “company vision/mission” ultimate goal for Blue Goji is to create a refined omnidirectional treadmill that aligns with the brand and targets rehabilitation applications by incorporating virtual reality (VR) as well as gaming features.

Team: Erik Takada [EECS], Pranav Veluri [EECS], Huachen Wang [ME], Megan Zhang [ME]
Advisor(s): Alice Agogino
Project ID: 67

Methane leak detection is a pertinent issue, as it is incredibly toxic to humans and a contributor to climate change. Because most detection methods require users to expose themselves to methane leaks, we aim to integrate various methane gas sensing technologies into a remote multimodal platform which will provide safe and accurate methane leak detection. Long range and contact sensors will be housed on individual droppable tensegrity robots to measure methane leaks in danger zones to maximize safety. Ultimately, the platform will utilize AI-integrated methane sensors for methane detection, and data collected from the robot array will be uploaded to the cloud for post-processing, interfacing, and storage using robust transformer and convolutional neural network models.

Team: Tarun S. Ganamur [ME], Masson Hung [ME], Edward Shi [ME]
Advisor(s): Eric (Yongkeun) Choi
Project ID: 92

Currently, most autonomous vehicles rely on data limited to their own system and sensors to operate. The development of connected autonomous vehicle (CAV) technology enables communication from traffic infrastructure and other vehicles to the CAV, which allows for safer and more energy efficient decision making on the road. Our team is creating a Model Predictive Controller based on an underlying vehicle energy consumption model, thereby maximizing vehicle energy efficiency. Our solution utilizes both simulation and real hardware on an off-the-lot EV to test and validate our model and controller.

Team: Cody Wiebe [ME], Kaan Beyduz [ME], Yiwei Jiang [ME], Haoyu Zhang [ME], Osaruese Asoro [ME]
Advisor(s): Mark Mueller
Project ID: 97

Due to Federal Aviation Administration (FAA) regulations, no fly-zone areas limit the feasibility of drone delivery due to limited reach of customers. To tackle this, our team is designing a mobile robot capable of transitioning between air and ground modes whilst using the same four motors to power both the propellers and wheels for ground mobility, reducing design complexity and increasing payload capacity.

Team: Yongqi Zheng [ME], Liam Parrish [ME], Ashna Reddy [ME]
Advisor(s): Doug Hutchings, Kingston Chua
Project ID: 68

To open new markets for Squishy Robotics’ technology, our team has developed a smaller tensegrity robot using model based design, and an early wildfire detection sensor payload using a customer centric approach to develop a product specification. By introducing more compact robots and new sensing applications, we can lower the risk associated with deploying sensors into hazardous situations.

Team: Ameera Elgonemy [ME], Sean Chu [ME], Brian Chung [ME], Dillon Balk [ME], Anusri Sreenath [EECS]
Advisor(s): Taylor Waddell, Hayden Taylor
Project ID: 14

Sending replacement parts or tools to astronauts in Earth’s orbit can be costly and time inefficient, while using traditional additive manufacturing technologies such as 3D printing or metal printing can take several hours to print inch-scale parts. SpaceCAL uses a relatively new additive manufacturing technique called Computed Axial Lithography (CAL) which uses projected light to print parts from resin in under a minute. The team is working to develop a payload that can automatically print and post-process four resin parts at once in microgravity aboard a Virgin Galactic spacecraft.

Team: Eric Tai [EECS], Heng Yu Lin [ME], Sizhe Tao [ME], Zaowei Dai [ME], Zhengrong Tang [ME]
Advisor(s): Tony Zheng
Project ID: 86

Painting a large mural can be extremely labor-intensive and time-consuming. The goal of this project is to develop an autonomous painting robot that can take an image input and
automatically paint it on a canvas. The robot consists of a robotic arm atop a mobile robotic
platform. An algorithm will determine the brush strokes necessary to recreate the image with minimal loss. It will then use camera sensors and vision algorithms to determine its location in space and actively adapt to variances from the desired brush strokes.

Team: Austin Marr [ME], Arya Goutam [ME], Yu-Han Wu [ME], Yining Wang [ME]
Advisor(s): Francesco Camozzi, Tarek Zohdi
Project ID: 126

By 2030, more than half of cars sold annually will be electric vehicles (EVs). Our team is developing a single-pedal interface for EVs to improve the driver’s control and comfort by combining Brembo’s advanced braking technology with a regenerative braking system. Our solution replaces passive hydraulics with cutting-edge sensors and intelligent haptic feedback improving the driver’s awareness, safety, and fun on the open road. 

Team: Robin Dumas [EECS], Mahir Rafi [ME], Po-Wen Huang [ME], Chung Chi Huang [ME], Tzuyi Yang [ME]
Advisor(s): Koushil Sreenath
Project ID: 120

Traditional static robotic arms that perform repetitive tasks (e.g. pick and place, organizing, sorting etc.) are greatly limited by their physical range. Team Robomotion aims to increase the reachable space of robotic arms by incorporating a mobile quadrupedal robot, with the ultimate goal of performing long-term tasks such as building a miniature brick castle. From a given castle blueprint (e.g. a 2D image), this autonomous vision-based robotic system determines and executes the required sequential steps to build the desired structure, requiring minimal user intervention.

Team: Encheng Liu [ME], Qingyue Liu [ME], Jiahao Zhao [ME], Huu-Quang Huynh [ME], Francisco Ramos [ME]
Advisor(s): Allen Yang
Project ID: 17

Our project focuses on creating an affordable and accessible autonomous go-kart by repurposing the hardware and chassis of conventional go-karts. This go-kart will serve as the foundation for an university-level autonomous racing competition, employing advanced technologies such as LiDAR, cameras, and GPS for trajectory planning and prediction, thus improving environment representation, perception, and localization. Through these combined efforts, our project aims to drive innovation, inspire collaboration, and pave the way for a future where autonomous vehicles revolutionize the world of motorsports.

Team: Haoting Feng [ME], Nan Liu [ME], Mingyu Zhu [ME], Dong Hyun Kim [ME], Su Sung Kim [ME], Shih-Wei Wang [ME]
Advisor(s): Homayoon Kazerooni
Project ID: 7

Warehouse workers commonly suffer back pain due to their duties. While exoskeletons could alleviate this issue, they face challenges in widespread adoption due to cost, weight, and manufacturability. We utilize a cable system to connect the hip and knee joints, applying supporting torque on both joints using one motor. Using statics structure and Lagrangian dynamics analysis, we calculate its output at each timestep of the squatting motion with heavy loads to provide balanced and adequate support. The proposed design creates a less expensive and lighter exoskeleton and hopefully provides more access to warehouse workers, benefitting their work conditions and well-being.

Team:Clea Rita Al Haddad [ME], Abhay Bhandari [ME], Isaiah Dillard [ME], Jiyun Hwang [ME], Hugo Pernet [ME], Haosen Sun [ME], Tong Wang [ME]
Advisor(s): Homayoon Kazerooni
Project ID: 6

Work Related Musculoskeletal Disorders (WRMDs) are among 30% of the most common workplace injuries and causes of absenteeism in manufacturing companies. Our capstone aims to design and develop an exoskeleton system to assist warehouse and factory workers during bending and over-the-shoulder tasks by supporting their shoulder and hip joints. To further accomplish the goal, our team will combine existing BackX and ShoulderX exoskeletons from SuitX company, coupled with a novel single actuator mechanism and power transmission system. This reduces the weight and cost of the exoskeleton and will prevent long-term injuries, increase productivity, and create an efficient workforce.

Team: Arman Baradaran [BIOE], Rajveer Oberoi [ME], Varin Kansal [ME]
Advisor(s): Kosa Goucher-Lambert, Ananya Nandy
Project ID: 154

Robotic automation is garnering more attention than ever, and its market is predicted to be worth $184.75 billion by 2030, but how much do engineers trust robots to perform complex tasks accurately and smoothly? The goal of our project is to gauge the psychological processes behind the confidence that human supervisors have in automated machines by tracking patterns of human intervention during sequences of robotic movement. To study such a nexus between human cognition and robotics engineering, we are using the Unity Engine to build an animation that incorporates simulated robotic assembly with agency for human intervention.