The two-hour event, free and open to the public, will feature Berkeley MEng and Fung Fellowship teams sharing the results of their team-based projects. This will be an interactive exposition. Teams will be showcasing their work either on a table top demo or poster format. This year we will be awarding Audience Appreciation Awards for both MEng and Fung Fellowship Projects! Light refreshments will be served.
- 4:30pm – Event Opens
- 4:40pm – Welcome remarks by student emcees, Thomas Guan, MEng ‘23 (ME) and Shreya Aviri FF ’24 (Conservation + Innovation)
- 4:45pm – Opening remarks by College of Engineering Dean Tsu-Jae King Liu and Fung Institute Founder Coleman Fung
- 4:55pm – Student Project Expo opens
- 6:20pm – Awards
- 6:25pm – Closing remarks by Fung Institute Faculty Director Prof. Anthony D. Joseph
- 6:30pm – Event Closes
- Reshape the Future. Come celebrate the different ways students across our Institute are solving challenges facing society through interdisciplinary teams and collaborating with community and industry partners.
- Networking. Attendees from across industry and campus communities will come together with Berkeley students, faculty, and alumni while learning more about the latest innovations and social impact.
- Industry-Wide Content. Attendees have the opportunity to explore trends across multiple fields and concentrations within various industries – healthcare, automotive, conservation, robotics, and more.
- Personalized Experience. Attendees can curate their event experience depending on interest. With more than 75 teams showcasing their work, there’s something for everyone!
2023 Award Winners
The Fung Institute Mission Award is awarded to the capstone team who has exemplified the mission of the institute “by solving the world’s problems through innovation, technology, and collaboration across boundaries.”
Winning Team: Generalizable Multi-Task Extraction of Breast Cancer Pathology
Team: Zhiwei Zheng, Ahmed Wali, Yan-Ning Yu, Yuwei Quan
Advisors: Madhumita Sushil (UCSF), Gabriel Gomes
The Capstone Technical Leadership Award is awarded to the MEng capstone team that most effectively demonstrates MEng Leadership principles: identifying an opportunity; generating a solution; including and convincing stakeholders of the proposed solution.
Winning Team: Control and Adaptation for Rehabilitation Exoskeleton Systems
Students: Shuaiwen Huang, Jessica Hsu, Michael Wang, Liuyu Wang, Yuqi Jin
Advisor: Homayoon Kazerooni (suitX)
The Capstone Innovation Award is awarded to the MEng capstone team who has most effectively demonstrated: a) the relevance of the problem they are trying to solve; b) the originality of their proposed solution; c) and the potential of their project’s impact.
Winning Team: 3D Cryoprinting
Students: Joseph Roux de Bézieux, Fiona Le Gaonac’h, Gakpe Mckenzie, Ania Boukhezna
Advisor: Boris Rubinsky
The MEng Showcase Audience Appreciation Award is awarded based on the live voting at the showcase event for the team who had the most engaging, inspiring, or impactful display:
Winning Team: Predicting Coma Recovery Using Massive Neurophysiology Datasets
Students: Josephine Daelemans, Ghita Houir Alami, Wassim Ben Sallah, Bronte Kolar, Sally Ahmed
Advisors: Edilberto Amorim (UCSF/Neurology), Gabriel Gomes
Master of Engineering Projects
Advanced Manufacturing and New Materials
Team: Issam Bourai [ME], Garrett Miller [ME], Shiv Makim [ME], Aemon Li [ME], Jake Nickel [EECS]
Advisor(s): Taylor Waddell
Project ID: 196
The SpaceCal team is working on developing and testing a novel volumetric additive manufacturing method designed to allow low-cost, precise and diverse parts to be printed in space. This method aims to help reduce the cost of space launches by reducing the amount of parts that need to be carried up in space while serving as a dynamic solution to help fix machines in space.
Team: Yue Li [MSE], Pin Ching Lin [ME], Haoyu Chang [ME]
Advisor(s): Jiachen Li, Kai Xu, Junqiao Wu
Project ID: 67B
With the technology of TARC (Temperature-adaptive radiative coating), temperature-controlling energy can be saved by altering material emittance. Our team is working on TARP (painting), an upgraded liquid form of TARC, which inherits the functionality of TARC and has a more abroad-applied situation. This liquid product is a more durable and user-friendly approach within fields like resident construction, air space optimization (satellite building), and even in the clothing industry.
Team: Colin Jensen [ME], Woojoo Lee [ME], Shun Chen [ME], Yuxuan Chen [IEOR], Asees Sarna [ME]
Advisor(s): Sebastien Lam, Peter Hosemann, Matthew Sherburne
Project ID: 102B
In the modern world, microfluidic devices are instrumental in almost every sphere of life, from automotives to medical devices. This diversified application of microfluidic devices has put a strain on the supply of microfluidic devices, due to their exclusive use of a rare metal called gallium. These concerns bring us to our objective, which is to investigate the feasibility of tin as an alternate molten metal for microfluidic devices and their applications. This adaption will lead to more diverse and sustainable applications in both the manufacturing and electronic industry.
Team: Yuxin Luo [MSE], Haowen Hu [MSE], Andika Nugroho [MSE]
Advisor(s): Zakaria Al. Balushi
Project ID: 120
The growing demand for semiconductor materials requires improved fabrication processes. Our team aims to optimize the slot-die coating technique, a trending coating method that is simpler, cheaper, and easier to scale up than traditional methods. However, achieving high-quality outcomes requires adjusting different parameters, including modifying ink recipes and coating parameters. We are using eco-friendly materials to adjust ink rheology and optimizing parameters such as dispense rate, coating speed, and slot-die head-to-substrate gap height. These efforts will allow us to create highly uniform and thin films suitable for semiconductor materials.
Team: Mona Zheng [BioE], Cole Ingamells [BioE], Lauren Mathis [BioE]
Advisor(s): J. Christopher Anderson
Project ID: 175B
Our team is optimizing a strain of self-lysing Escherichia coli to be harnessed in a nutrient recycling system for biomanufacturing applications. Genetic engineering techniques have been utilized to identify lytic proteins and integrate inducible expression constructs. Induction of our genetic construct leads to the degradation of the plasma membrane and releases the nutrients contained within. These nutrients can be utilized as future feedstock, thereby making future cell growth cheaper and greener.
Artificial Intelligence, Machine Learning, and Data Science
Team: Shruti Satrawada [EECS], Sivani Voruganti [EECS], Victor Li [EECS], Zhen Jiang [EECS], Gen Yang [EECS]
Advisor(s): Anant Sahai
Project ID: 70
Wireless Spectrum data powers the communication around us—from phones to WiFi. The data’s complex and dynamic nature requires a custom pipeline to handle the massive volume while ensuring quality and reliability. To allow the public to utilize this data we are building an end-to-end data pipeline that collects, stores, visualizes and analyzes it. Our unique user-centric approach allows the public to harness this data by replicating our pipeline with start-from-scratch tutorials. Our pipeline utilizes the NATS.IO system for data transmission, an Elasticsearch database for AI/ML application integration, and Prometheus for status monitoring, all within a distributed AWS-cloud-based environment.
Team: Adheesh Shenoy [EECS], Pranjal Sinha [ME], Yixuan Tao [IEOR], Mitchell Ding [EECS], Pranav Chopada [EECS]
Advisor(s): Harrison Shaw, Ashby Monk, Tarek Zohdi
Project ID: 110B
Shelton AI advises Sovereign Wealth Funds on assets allocation through quantitative analysis. The objective of this capstone project is to use NLP to identity key attributes in financial documents. The process involves annotating the data, augmenting the data, and training and testing a NER model
Team: Neil Palleti [EECS], Yukun Zheng [EECS]
Advisor(s): Michael Mahoney, Yaoqing Yang, Benjamin Erichson
Project ID: 186
Recent developments in deep learning have greatly increased the power of machine learning models, but they are vulnerable to backdoor attacks. These are malicious attacks used to fool a model and cause it to behave incorrectly and dangerously. Our project uses weight analytics to study which model training methods and hyperparameters lead to the best backdoor detection algorithms. Our results serve as guidance when building models regarding designing the best training strategies in the presence of backdoor attacks.
Team: Ruben Partouche [CEE], Mathieu Dupont-Rio [CEE], Lea Schapira [IEOR]
Advisor(s): Omar Khan, Jackson Mills
Project ID: 187
The purpose of the project is to tackle the accountability problem in crowdfunding using blockchain and smart contract technology. The project aims to increase transparency, security and decentralization of the crowdfunding process and involves developing a crowdfunding platform using Ripple’s blockchain technology, where creators can create campaigns with milestones, and backers can contribute to the campaign. The use of such technologies are allowing the crowdfunding process to be more reliable and trustworthy.
Team: Edwin Joseph [ME], Zhenyi Yue [ME], Liberty Hudson [ME], Sunny Chu [EECS]
Advisor(s): Aaron Gyure, Sean Houlihan
Project ID: 216
Checkout lines are the largest cost to grocery delivery providers who employ over 800,000 drivers in the United States alone. The kwikkart team is creating a smart checkout solution for delivery providers to make grocery delivery faster, cheaper, and more widespread. Our team integrates artificial intelligence and computer vision into custom hardware to provide a scalable and instantly deployable checkout alternative.
Team: Julien Raffy [IEOR], Ming Martin Liu [CEE], Yizhe Zhao [EECS], Zhuofan Li [EECS]
Advisor(s): Lee Fleming
Project ID: 77
Engineers typically do not read scientific research papers although it represents a pillar of innovation. They often rely upon gatekeepers who are people involved in both patents and scientific papers hence bridging the worlds of theory and application. However, there is currently no publicly available data on science gatekeepers. We are leveraging machine learning methods such as NLP and Random Forests models to identify these important figures of the scientific field. This is an opportunity to boost innovation and entrepreneurship by connecting promising climate technology to its contributors and potential investors.
Team: Allen Pan [EECS], Sohum Desai [EECS], Nathania Santoso [EECS], Yinuo Wang [EECS], Yuqi He [EECS], Cecil Symes [EECS], Winston Sun [EECS]
Advisor(s): Mohamed Ibrahim, Jan Rabaey
Project ID: 95
Artificial intelligence (AI) has become ubiquitous in today’s world and has revolutionized many industries; however, it can be expensive in terms of energy and resources. To address these limitations, our team is designing a hardware accelerator–a specialized chip–for hyper-dimensional computing (HDC), an emerging brain-inspired approach for AI applications. This accelerator will work in tandem with a central processing unit (CPU); together, this entire chip will provide a platform with improved resource demand and power consumption in comparison to the existing state of the art methods.
Team: Wesley Cheung [IEOR], Hongfei Guo [IEOR], Yuqing Xu [IEOR], Menghan Chen [IEOR]
Advisor(s): Samuel Gonzalez, Ivonne Avila, Lee Fleming
Project ID: 126
For model quality control at MSCI Inc., an investigation process, currently performed by subject-matter experts, is conducted to provide explanations for the anomalies. By exploring approaches to link output changes to root causes, we develop a Python-powered Auto-Investigation Engine that automatically verifies and explains output anomalies with minimal human intervention. Compared with a manual procedure, our engine qualifies for a faster and more robust investigation process, addressing the rising issue of scalability brought about by the increasing amount of data from market activities.
Team: Hirotaka Ishihara [EECS], Zhengjiang He [EECS]
Advisor(s): Professor Dawn Song, Ph.D. Xiaoyuan Liu
Project ID: 131
Our project is to create a decentralized collaboration marketplace that enables data owners to list, sell, and work together with data consumers in one unified platform. By using blockchain technology and crafted cryptographic features, our marketplace will promote a more conducive environment for collaboration. With our platform, the data owner can monetize their data without worrying about privacy, loss of ownership, or fraud.
Team: Han Feng [CEE], Gabriel Dolique [CEE], Yidi Xu [IEOR], Yike Chen[IEOR]
Advisor(s): Zeyu Zheng
Project ID: 138A
Customer satisfaction and customer reviews are critical for the logistics industry, including app-based services, e-commerce, package, and platforms that easily incur long waiting times. The project aims to develop a new perspective on customer satisfaction analysis using machine learning techniques. Through analysis based on waiting times, merchandising data, etc., it is expected that insights can be gained into service improvements in the logistics industry, such as what is a better way to allocate inventory, quantitative analysis of the importance of service satisfaction to customers, etc
Team: Mathew Martel [EECS], Yuxin Hu [ME], William Verstraete [ME], JiaYang Li [MSE], Chia-Yang Fang [NE]
Advisor(s): Prof. Peter Hosemann, Christopher Reis
Project ID: 154
Superconducting magnets are fundamental to creating the extremely high temperatures and pressures that nuclear fusion reactions require. Our goal is to understand and quantify the changes in these materials as they are bombarded by neutrons during reactor operation which might lead to the loss of the material’s superconductivity. We plan on doing this by training a machine learning algorithm on nanoscale material images taken by a transmission electron microscope in order to discover the limitations of the materials and advance the technology.
Team: Yiting Gan [IEOR], Yannan Niu [IEOR], Sining Shen [IEOR], Xiaojian Li [IEOR], Anqi Fan [IEOR], Solar Shao [IEOR], Lingxiao Pan[ME], Fan Fei [IEOR]
Advisor(s): Victor Detavernier, Matthew Kanter
Project ID: 162A and 162B
Furiends is a mobile game where users adopt virtual dogs and are incentivized to be physically active. The project team plans to use machine learning to analyze user data and develop a user stickiness strategy to enhance the game’s design and functionality. The goal is to motivate users to engage with the game and keep them coming back through personalized rewards and social features.
Team: Maria Albery [IEOR], Yihua Huang [IEOR], Zhiyang Liang [IEOR], Lijie Liu [IEOR], Yutong Yang [IEOR]
Advisor(s): Xin Guo, Lizeng Zhang
Project ID: 224D1
Our project utilizes reinforcement learning (RL) to aid in the stock-investing decision making process, where the investment agent learns to synthesize market information to optimally choose when to buy, sell, or hold a stock. Our approach is unique because of the technical indicators we implement to help in the agent’s decision making process. The agent is trained with several different RL algorithms, and then is tested against other market participants. The RL agent outperform naive agents such as a random-action agent, and can also outperform an intelligent machine learning agent.
Team: Charlotte Jin[IEOR], Jingqi Ma[IEOR], Ethan Choukroun[IEOR], Haoyang Wang[IEOR]
Advisor(s): Xin Guo
Project ID: 224D2
Industry practitioners as well as academic researchers have been proposing optimal trading strategies to make money in the stock market for years. However, complex and unknown market dynamics pose significant challenges on the development and evaluation of the strategies. Our team is developing an optimal trading strategy in a market simulation environment that mimics the real-world market to account for the constantly changing market variables. We train reinforcement learning trading agents in the Agent-Based Interactive Discrete Event Simulator (ABIDES) to maximize daily investment profits. The proposed method will greatly facilitate investment firms making more profitable trading decisions.
Augmented and Virtual Reality
Team: Colin Yang [ME], Julia Tanner [ME], Yunchen Wu [ME]
Advisor(s): Liwei Lin
Project ID: 86A
Have you ever wanted to pet a dragon or hold a lightsaber? Meta’s Oculus Quest 2 brings the user into a virtual world by utilizing vision and hearing, but it does not incorporate the sense of touch. Our team is working to replace these bulky handheld controllers with a haptic glove that features flexible sensors and motors to emulate physical interaction with VR objects. The result is a cheap, safe, and user-friendly add on that provides a more immersive experience.
Team: Fangyi Lu [BioE], Junkai Shao [ME], Yujie Zhang [ME]
Advisor(s): Liwei Lin
Project ID: 86B
Currently, about 21% of people in the United States have used or purchased VR devices. However, these VR devices are monotonous and uncomfortable for most people to wear due to the large area of contact with their skin. Our team is working on designing a comfortable exoskeleton glove with multiple non-contact haptic feedback on fingertips. The device incorporates ultrasonic transducers to generate and tune the non-contact sensations at low voltage to improve the user experience and create a realistic feedback system that will facilitate the VR industry to expand its applications.
Team: Sean Wang [EECS], Chuanyu Pan[EECS]
Advisor(s): Allen Yang
Project ID: 91A
Emerging technologies such as autonomous driving and augmented reality (AR) suffer from a lack of fundamental support for the detection and tracking of 3D objects, thereby limiting their overall usability. The implementation of high-precision and real-time 3D object tracking fortifies the stability and resilience of autonomous vehicles, while also expanding the possibilities for AR applications. In this work, we design a 3D object tracking framework and is applied to autonomous racing and augmented reality applications.
Team: Joshua Duarte [ME], Prachi Anand Tappu [ME], Alex Rivero [ME], Pavan Reddy [ME], Sebastian Harper [ME], Ethan Chung [BioE], Jiahua Yang [ME], Jia-You “Billy” Lin [ME], Siddhant Kaushik [ME], Bryan Tran [ME]
Advisor(s): Matthew Kanter, Tiffany Tao, Victor Detavernier
Project ID: 146
The goal of the project is to improve the industrial design and real-time controls of a scaled-down, omni-directional treadmill paired with VR. This project is iterative with this year’s team having the goal of creating a brand new prototype with reduced height, noise, and overall footprint. The team also plans to implement a new controls system that keeps the user centered on the treadmill at all times. The purpose of the treadmill is to enhance physical therapy and rehabilitation in patients afflicted with Alzheimer’s, Parkinson’s, or recovering from a stroke.
Team: Joey Hou [EECS], Lu Li [BioE], Tianjian Xu [EECS]
Advisor(s): Prof. Brian A. Barsky
Project ID: 181B
Over 2.2 billion adults in the world need to wear eyeglasses. They need to either keep their glasses on when using virtual reality (VR) headsets or purchase prescription lenses for their headsets, which could be uncomfortable or expensive. There needs to be a lightweight, affordable, and universally applicable way for such users to use VR headsets without eyeglasses. We applied the vision correction display technology to VR headsets in order to allow people with vision problems to enjoy an eyeglasses-free VR experience in significantly high resolution and contrast. Our solution combines the prefiltering algorithm and pinhole masks, which can be easily customized and manufactured. The algorithm first processes images on the display screen. Then the light from the screen goes into the user’s eyes through a pinhole mask, which consists of tiny grid-shaped holes and filters light. The user can then view clear images directly on VR headsets by looking through pinhole masks over the prefiltered picture. This solution could improve the user’s sightedness in immersive reality with a lightweight and at a low cost.
Team: Ruizhe Jin [ME], Xiao Dong [ME], Yiru Mao [ME]
Advisor(s): Liwei Lin
Project ID: 86C
Our project aims to design and develop haptic control sensors that can be implemented on an individual’s glove with sensory feedback control. These sensors can be used on future augmented and virtual reality systems, such as goggles and navigation systems. The project team has focused on improving the fabrication technique, signal processing technology, and machine learning algorithm to achieve high resolution and good stability in wearable sensing devices.
Energy and the Environment
Team: MaoFeng Lin [ME], Suda Xu[MSE], Emmanuelle “Manou” Samama [MSE], Humza Habib Sadiq [ME]
Advisor(s): Peter Hoseman, Matthew P. Sherburne
Project ID: 78A
This project aims to recycle degraded solar panels by extracting and processing their silicon content for use in alternative industries such as metallurgy. We will evaluate various techniques for economic and ecological feasibility and compare their environmental impact. Our goal is to promote a circular economy and reduce waste while reusing valuable resources.
Team: Ryan Ho [ME], Thomas Guan [ME], Lin Gan [ME]
Advisor(s): Doug Hutchings, Alice Agogino
Project ID: 157
The Squishy Robotics Team developed a novel robotic payload packaged in a lightweight, drone-droppable tensegrity structure that allows for remote deployment and continuous multipoint surveillance of hazardous areas.
Team: Jorge Elvira [ME], Umar Shaheed [ME], Catherine Robbins [ME]
Advisor(s): Reza Alam
Project ID: 215
Research into using inflatables for renewable wave energy generation. This involves lab-scale testing, rapid prototyping, and design. We have made significant progress in our energy generation capacity.
Team: Daksh Aggarwal [ME], Luca Aringsmann [ME], Wuwei Mo [MSE], Lauren Takata [ME]
Advisor(s): Dr. Christine Spiegelberg (Siemens Energy), Dr. Tarek Zodhi (UC Berkeley)
Project ID: 195
Investing in solutions to reduce global carbon dioxide levels is crucial to combat climate change. Direct Air Capture (DAC) is a promising solution to achieve this goal. However, current efficiencies are low and largely depend on sorbent, or filter material, selection. Therefore, our team is conducting a techno-economic analysis of conventional and emerging solid sorbents while also considering overall environmental impacts. We look to quantify data and create a comprehensive model to compare sorbents at a large scale in order to solve this multi-objective optimization problem. We hope to provide novel insights into the current labyrinth of DAC and sorbent technologies..
Health and Well-Being
Team: Liuyu Wang [ME], Jessica Hsu [ME], Zhixiang Wang [ME], Shuaiwen Huang [ME], Yuqi Jin [ME]
Advisor(s): Homayoon Kazerooni
Project ID: 71B
Our capstone aims to design and develop an exoskeleton system to assist warehouse workers or heavy-lifting workers during squats and other similar movements by supporting their hips and knees. To further accomplish the goal, our team will utilize the existing SuitX technology, coupled with a novel linkage mechanism to reduce the weight and cost of the exoskeleton. This will result in a more injury-prone, productive, and efficient workforce.
Team: Phoenix Ding (BioE), Karthik Raj (BioE), Dhanush Nanjunda Reddy (EECS)
Advisor(s): David Wu
Project ID: 87
Current average wait time for dermatologist appointments is around a month, and the disparity in dermatology access between rural and urban areas intensifies this issue. Teledermatology has the potential to rectify these issues, but an inability to accurately and consistently document skin conditions prevents it from doing so. Our team has developed an iOS app that creates 3D models from videos of skin using photogrammetry software that is deployed on the cloud. This solution will provide the necessary depth perception for dermatologists to accurately evaluate skin conditions and increase overall access to dermatologic care.
Team: Jethro Cho [BioE], Maya Girimaji [BioE], Christian Mast [ME], Shiru Shong [IEOR], Roy Yang [ME]
Advisor(s): Gordon Lee, Dr. Jeff Suh, Dr. Grace O’Connell
Project ID: 116
Nasal allergy problems affect nearly 10% of all adults and children, with many of them relying on nasal irrigation to relieve symptoms. We are creating a novel nasal rinse device that decreases reoccurring infection rates by improving the sanitization process. Our product features an ergonomic bottle design with drying and UV-C sterilization capabilities. By improving hygienics in the rinsing process, this innovative solution will improve patients’ management of sinus infections.
Team: Kieran Weiszmann [ME], Gabrielle Evans [BE], Madeleine Gorham [BE], Boxuan Song [ME], Miguel Montalban [ME], Andy Yau [BE]
Advisor(s): Elise Scipioni, Tiffany Tao, Ryan Kaveh
Project ID: 147A and 147B
The measurement of cognitive activity can improve clinical outcomes from physical rehabilitation. This can be achieved through electroencephalography (EEG), which is the measurement of electrical brainwave activity. However, this technology is not well developed for wearable and user-friendly contexts, and the limited existing wearables are not suited for data collection in motion.
Our goal was to develop a functional in-ear EEG device to collect real-time cognitive data during physical rehabilitative tasks. Our solution utilized EEG electrodes strategically placed on an earpiece to record data, along with a paired headset that maintains earpiece stability while housing a custom module for real-time data processing and transfer for visualization.
Team: Natalie Franklin [BioE], Tian Xia [MSE]
Advisor(s): Syed Hossainy, Dorian Liepmann
Project ID: 169
Our team is designing a heart patch using piezoelectric material for heart attack treatment. We are using piezoelectric material to convert muscle contractions from the heart into voltage, which will stimulate cardiomyocyte maturation and differentiation and regenerate the dead tissue. The patch will be placed on the infarcted area of the heart. We are developing films from multiple material combinations. To test our films we created a test platform and we are applying a force at a rate that mimics the heart, and measuring the voltage output.
Team: Jason Torres [ME], Johanna Bustamante [ME], Zehao [Guo]
Advisor(s): Celeste Castillo
Project ID: 176
The team aims to create a portable, modular, and cost-effective contrast therapy device using a flexible Peltier unit, improving user experience and compliance while providing enhanced muscle and soft tissue recovery benefits.
Team: Irina Hallinan [EECS], Jinmeng Zhang [EECS], Isadora Smith [BioE]
Advisor(s): Professor Brian Barsky
Project ID: 183B
In the US, over 500,000 people who are deaf or hard of hearing rely on American Sign Language (ASL) to communicate. Currently, people who communicate in ASL hire live translators to participate in virtual meetings, which is time-consuming and expensive. Sign Language Assistant for Meetings (SLAM) bridges the communication gap between hearing and non-hearing people. We employ a Human-Centered Design approach to develop a prototype for an automatic ASL-to-text translation for Zoom meetings. Our solution generates the translation of signs in real-time, using machine learning to detect key-points on the user’s hands, face, and body from the computer camera.
Team: Gavin Chan [EECS], Bill Linzhe Tong [EECS], Jinghua Zhang [EECS]
Advisor(s): Brian Barsky
Project ID: 183C
While keyboards are the primary text input method for computers, individuals with hand and finger dexterity issues may find them difficult to use, such as patients suffering from arthritis, stroke, old age, or repetitive stress. Our team is creating software to capture hand movement and characters drawn by fingertips to simulate keyboard typing, eliminating the need to press individual keys with fingers. By using computer vision and machine learning to capture and recognize air-drawn characters, the result is an accessible method of keyboard input that incorporates existing webcams.
Team: Brian Hong [BioE], Jackson Wagner [EECS], Zhiqi Yan [EECS], Xu Zhao [ME]
Advisor(s): Brian Barsky
Project ID: 183D
We have created a program to control a computer using alternative methods for paralyzed users. Our program utilizes a webcam and microphone built into a computer as input. By measuring the user’s facial landmarks, estimating the head pose, and processing audio inputs, we provide a low cost solution to control a personal computer.
Team: Phong Ha [EECS], Shaopu Song [EECS], Mengqi Huang [IEOR], Tanel Petelot [ME]
Advisor(s): Greg Tikhomirov
Project ID: 112
We propose to develop an AI-assisted algorithm to communicate with your deceased family members by training a virtual assistant on data offered by the users. This will not replace your close one, but it may help you alleviate the transition after their loss. We achieve this by fine-tuning a pre-trained large language model such as GPT-3 on personal data, including but not limited to, text messages, emails, and tweets.
Team: Josephine Daelemans [EECS], Ghita Houir Alami [IEOR], Wassim Ben Sallah [BioE], Bronte Kolar [BioE], Sally Ahmed [BioE]
Advisor(s): Edilberto Amorim
Project ID: 141
Predicting coma recovery is a complex process that requires careful analysis of patient data and medical expertise. We are developing methods that integrate time-series electronic health record data to uncover new insights about the mechanisms of neurological recovery after acute brain jury. Our models rely on Convolutional Neural Networks and Long Short Term Memory, which integrates an additional network architecture that makes the “black box” model interpretable.
Team: Zhiwei Zheng [EECS], Ahmed Wali[BioE], Yan Ning Yu [BioE], Yuwei Quan[IEOR]
Advisor(s): Madhumita Sushil
Project ID: 151
Cancer pathology contains a lot of information about cancer diagnosis and tumor characteristics which contribute to cancer diagnosis, treatment planning, and research. However, manual extraction of pathology information is an inefficient and error-prone process to get targeted pathology parameters. As a result, we aim to develop multi-task machine learning models to help extract information in a more efficient and scalable way. With our models, doctors and researchers will be able to access breast cancer pathology information more efficiently and with high classification accuracy and generalizability.
Team: Gongao Xue [EECS], Han Zhang [BIO], Chenyan Zhang [EECS]
Advisor(s): John Anderson
Project ID: 174
Despite the importance of enzymatic pathways in biological research, there lacks an organized system for accurate prediction of enzymatic reaction products. In this project, we design an improved algorithmic system that aims to abstract the existing enzyme data for further prediction of unknown ones. The implementation is in Java and Python, utilizing the chemical classes from the Rdkit library. Our group uses pathways to abstract the structure of molecules and ML-like algorithms to make predictions.
Team: Elijah Grimaldi [NE], Oscar Matousek [NE]
Advisor(s): Kai Vetter, Woon-Seng Choong
Project ID: 219
Proton cancer therapies have become popular by allowing for the accurate targeting of tumors and avoiding damage to healthy tissue, causing fewer side effects compared to other treatments. Understanding the range of dose distributions is vital to ensuring the efficacy and safety of the treatments – however, current methods have insufficient resolutions. ProCanTh is collimating characteristic prompt gamma rays onto a detector array, which are then used to reconstruct 2D images of the proton dose. This novel approach will determine the dose distribution to an accuracy of 1 mm – much smaller than typical cancerous tumors.
Robotics and Automotive Advancements
Team: Dahany Choi [ME], Paul Diaz [ME], Elyse Scalia [ME], Xinglong Li [ME]
Advisor(s): Homayoon Kazerooni
Project ID: 72B
BackX is a trunk-support exoskeleton that reduces the risk of work-related back injury by supporting the torso during lifting tasks. Our solution builds upon an existing backX design to make a device that is cheaper, lighter, easier to manufacture, and one-size-fits-all. This addresses the growing health issue while furthering exoskeleton technology in hopes of making these preventive health devices more accessible.
Team: Yuhang Peng [ME], Rafael Rivero [ME], Brady Johnson [ME], Hao Ouyang [ME], Jayant Kumar [ME]
Advisor(s): Xu Shen, Francesco Borrelli, Thomas Fork
Project ID: 75
Our team is developing an off-road rover capable of navigating through a solar farm autonomously. The objective of our product is to reduce labor requirements and construction timelines. To achieve this objective, our rover interfaces with onboard perception and control systems. These systems enable the rover to perceive its environment and determine its current position and velocity. Integrating these systems allows our rover to parallel park, avoid obstacles, and follow paths through the solar farm.
Team: Aditya Gupta [ME], Yibo Sun [ME], Steed Amegbor [ME]
Advisor(s): Eunhyek Joa
Project ID: 88
Our project consists of implementing the vision based localization approach in real time for an autonomous vehicle in urban road scenarios. Our solution incorporates the use of stereo camera calibration, mapping, image processing, image retrieval, and particle filtering to estimate the vehicle’s pose to a high accuracy. We aim to provide a relatively cheap solution by solely using stereo cameras compared to industry hardware (example. – LIDAR) for localization.
Team: Haoda Li [EECS], Yunhao Liu [EECS], Piaoyuan Yi [EECS]
Advisor(s): Avideh Zakhor
Project ID: 94
By leveraging traditional 3D reconstruction methods and deep learning, we solve the challenges and create an end-to-end system for for scanning indoor environments with commercially available drones. Our system can easily create accurate and high-quality 3D models for building interiors. The models are compatible with common 3D viewing and editing software. Our system is valuable for interior design, building inspections, and virtual tours.
Team: Rushawn Childers [ME], Karl Skeel [ME], Jiachen Su [ME]
Advisor(s): Laura Treers, Hannah Stuart
Project ID: 117
Our team is engineering an electromechanical model of the Pacific Mole Crab. We have been working on redesigning a previous prototype by shrinking its cross-sectional area, and adding more onboard motors. With four independently driven legs, our prototype will allow for advanced control schemes that will lead to operations such as locomotion, burrowing, and anchoring in the sand.
Team: Alex Bartoletti [ME], Alberto Villaverde Caridad [ME], Archit Srivastava [ME]
Advisor(s): Scott Ziegler
Project ID: 127A
With the 2025 NASA Artemis mission fast approaching, the world is in need of a reliable, streamlined, and energy efficient lunar resource transportation system. Our group’s goal is to design and test a reusable, rugged-capsule that can withstand high loads and extreme operating temperatures ranging from -200ºC to 100ºC. While delivering vital liquid payloads such as liquid oxygen (LOX) and water, Space Kinetic will establish a simple network to connect hard to access and hazardous locations with lunar labs and communities with minimal human involvement and energy consumption.
Team: Lucas Garcia [ME], Harper Xu [ME], Jessica Armstrong [ME]
Advisor(s): Scott Ziegler, Tarek Zohdi
Project ID: 127B
In 2025, the Artemis mission will send humans back to the moon with the hope of settling a permanent colony. But once there, building, maintaining and supplying the infrastructures (mines, factories, houses…) at the same location is challenging. Our goal is to design a capsule that transports solid cargos (like regolith, batteries or i-beams) via a catapult-like launching system. Using inflatable bags, the capsule will carry objects with different geometries and protect them from high acceleration. Our solution will allow fast and efficient long-range moon supply-chain.
Team: Rithin Venkatesh [ME], Yuxuan Guo [ME], Vighnesh Rane [ME]
Advisor(s): Doug Hutchings, Alice Agogino
Project ID: 158
Each year, the US Fire department deals with nearly 500,000 calls involving hazardous materials. To minimize first-responder hazmat exposure, mobile tensegrity robots with sensors can be deployed to assess hazardous situations. Current tensegrity robot navigation methods are complex, time-consuming, and require constant adjustment by trained individuals. These negatives hinder the adoption of tensegrity robots as first responders prefer easy-to-use, quickly deployable technology. Our goal is to create a user-friendly way for first responders to control mobile tensegrity robots through hazardous environments.
Team: Cheng-Kai Chen [EECS], Du Xiang [EECS], Constantin El Ghazi [ME], Hendrik Chiche [ME]
Advisor(s): Lou Graniou, Michael Kellman, Gabriel Gomes
Project ID: 163
As a goal to improve our data quality and enhance sensor fusion, our team decided to focus our efforts on algorithms to label radar data. Firstly, we performed object detection and annotations on image data using the YOLO model. We were able to accurately label and identify objects in the image and use the results as the labels for radar data. Secondly, we studied the camera projection and applied the knowledge to map radar data to be in the same space as our image labels. Lastly, we can apply clustering algorithms to automatically label radar data with our image labels.
Team: Chet Kruse [ME], Hanwen Wang [ME], Hunter Ragantesi [ME], Jiaming Chen [ME], Julien Rineau [ME]
Advisor(s): Bike Zhang, Shuxiao Chen, Koushil Sreenath
Project ID: 178
Legged locomotion offers robotics as a unique solution to applications such as search and rescue, and other industrial tasks. Complex terrain and dynamics environments pose challenges to developing an effective and robust legged control strategy. Our project focuses on enhancing the performance and agility of legged robots through combining the strengths of Model Predictive Control (MPC) and Reinforcement Learning (RL). Utilizing a combined Residual Learning strategy we aim to improve upon the limitations of classical predictive controllers and purely RL-based controls.
Team: Xiaofeng Zhao[ME], Zhiyuan Zeng[ME], Chenhang Yuan[ME],Tingyu Guo[ME]
Advisor(s): Koushil Sreenath
Project ID: 179A
While the proportional-integral-derivative (PID) controller has been the most basic and popular control technique for industry, it has the weakness of not being able to always perform optimally with constraint and disturbance. Therefore, advanced control techniques are needed for safe and efficient operations in complex environments. This portfolio will show two methods of learning MPC and neural networks to control aerial robots in two different application scenarios.
Team: Vincent Chendrawan [ME], Cole Whatley [ME], Antoine Terraux [ME], Honghui Xie [ME]
Advisor(s): Koushil Sreenath
Project ID: 179B
In 2021, an estimated 60,750 firefighters suffered injuries in the line of duty, 32% of which happened while fighting fires. To help mitigate the risk of injury involved with fire response, our team is working on a multi-drone system which can be used to carry and direct a hose. Our objective is to maneuver this two drone system using one simple, easy to use control scheme from a nearby device such as a laptop. The system is intended to eventually be applied to a running firehose at emergency sites to remove firefighters from the fire, making their job safer.
Team: Ryan Wang [ME], Stephanie Popielarz [ME], Coby Lim [CEE], Aneesh Jois [ME]
Advisor(s): Reza Alam
Project ID: 211
Shipping accounts for 3% of global greenhouse gas emissions; one cruise ship produces as much sulfur dioxide as 3.6 million cars, and as much NOx as 1 million cars. We are designing a swarm of low-cost autonomous ocean drones to collect wave data for an advanced routing algorithm. These ship routing algorithms can reduce ship fuel consumption, and thereby emissions, by more than 20%.
Team: Alex Sedov [ME], Pawandeep Dhall [ME], Siyuan Ren [ME], Christopher Wu [ME]
Advisor(s): Vu Vuong
Project ID: 218
The aim of the project is to develop a novel, autonomous lunar rover with high maneuverability on unpredictable lunar terrain, high scalability for ease of reproduction on any scale, and high modularity to meet a range of commercial needs. This rover will act as a versatile platform and offer advanced robotic solutions and services for frontier applications in space exploration.
Team: Zhongpu Diao [EECS], Shixin Yan [ME], PinYun Hung [ME]
Advisor(s): Murat Arcak, Alex Devonport
Project ID: 65
According to Forbes News, over 35,000 fatal car accidents occurred on roadways across the nation during 2022, another 1,600,000 crashes resulted in injuries, and 3,600,000 crashes caused property damage. Targeting such catastrophic results and social trends, Autonomous Defensive Driving Team is designing an autonomous driving technology that prioritizes safety and target following with real-time feasible vehicle control so that our technology can help reduce traffic accidents. Our solution incorporates sensors using LIDAR, control algorithms with MPC in convex sets, and state estimation with Kalman Filters, which achieves autonomous driving purposes in short reaction and precision.
Team: Lincoln Too [IEOR], William Delbegue [ME], Shuhao Lu [ME], Joyce Li [ME]
Advisor(s): Janelle Lines, Tarek Zohdi
Project ID: 180
In the United States, gas leaks contribute to approximately 5 billion dollars worth of loss per year. Our team is designing a drone with an image-based machine-learning algorithm and sensors to autonomously locate gas leaks. The machine-learning model relies on convolution neural networks (CNN). The final product will be an integrated solution that can improve the safety, efficiency, and cost of gas leak detection.
Team: Travis Welch [ME], Sangmin Sung [ME]
Advisor(s): Homayoon Kazerooni
Project ID: 210
The ShieldX, one of SuitX’s many exoskeleton products, has two crucial problems that hinder user’s comfort; Its rigid frame doesn’t have flexibility to comfortably bend forward; Its lumbar support feature pulls user’s shirt up when the user bends forward. Using human centered design techniques and design iteration, the team focused on developing the passive mechanisms necessary to solve the problem of transferring external loads from the user’s shoulders down to their hips for industrial workers.
Team: Hsiu-Ting Yeh [EECS], Chris Dong [EECS]
Advisor(s): Sophia Shao, Dima Nikiforov
Project ID: 214
Assessing how well robotic SoC perform certain tasks is difficult because of the way they interact with their surroundings and the complex software they use. To make this easier, we made a system that lets us test robotic SoC before they are even built. This system helps us see how well a robotic SoC can do a task from beginning to end by simulating the robot’s interactions with its surroundings.
Team: Izcalli Rios-Aguirre [ME], Arthur Nguyen [ME], Cristina Martinez de Juan [ME]
Advisor(s): Mah Kadidia Konate
Project ID: 217
Being the chosen hairstyle for millions of people around the world, hair braiding is a highly time-consuming and pain-inducing process with little evolution over thousands of years. HairRobotics takes inspiration from the textile and beauty industry to design a non-invasive and client-friendly device to assist hair salon workers. Our solution incorporates a device that retrieve and controllably release the lock of hair, into a gear-driven maypole braider creating three-section braids quickly, relatively painlessly, and easily.
We cannot guarantee that all guests attending the event have been vaccinated. Please review current UC Berkeley COVID-19 policies and resources. This information may be updated at any time and changes to local and state health regulations may alter event plans.
If you require an accommodation for effective communication (ASL interpreting/CART captioning, alternative media formats, etc.), please reach out to Jené Madison at email@example.com by Friday, April 28.