Nicholas Wen
Advanced Idea Mechanic currently located in the United States.
Open to new work opportunities!
Who Am I?
Hey, I'm Nicholas! Welcome to my spot on the web for my projects, resume, and personal thoughts.I recently graduated from Northeastern University with my M.S. in Robotics. At Northeastern my research was mostly focused on generating adversarial images to test neural network robustness. I completed my B.S. in Computer Engineering at University of California, Santa Cruz where I was a math, computer science, and physics tutor.I am often asked why I use the title of "Advanced Idea Mechanic". The title is a statement of my engineering philosophy. In my mind, to be an engineer is to take on the role of a translator, moving from "Advanced Ideas" into real world products, thus an "Advanced Idea Mechanic" is someone who develops, maintains, and improves upon the implementation of a high level "Advanced Idea". To describe myself as an "Advanced Idea Mechanic" is a statement about the end goal of engineering not to be the engineering itself, but rather on the impact on the world around us.To hear about some of my work, please visit the projects section!
Selected Projects
Breathing Easy with X-Ray Vision
: Detecting Pneumonia with SEResNets
Pneumonia is a vicious disease that kills almost 50,000 people in the United States alone. A chest X-ray (CXR) is the most common way to diagnose pneumonia, however identifying the disease can be very difficult due to the high visual similarity between pathologies. While work has been done to leverage convectional neural network models for pneumonia detection, some initial findings are uncorroborated. In this project, we investigate how true these claims are and discover if this technology is ready for wide-spread hospital use.
Implemented and verified findings of SEResNet with practical size and computational restraints for a hospital setting
Performed ablation study to observe effectiveness of 4+ large deep network models to detect pneumonia in chest x-rays
MIAH- Natural Language Processing Powered Assistive Feeding Arm
This project is motivated to propose an optimized voice-controlled robot to feed people with malfunctions of the nervous system that can cause a lack of intended movement or an excess of involuntary movement. We take advantage of natural language processing (NLP) to command a UR3 robot to find the user’s position in the robot’s workspace. Once the final destination is determined, an optimized trajectory is generated. In this work, the desired trajectory maintains the end effector rotation to avoid comestible spill.
Developed algorithms to interpret microphone inputs into arm movement frames using Google Speech to Text and Python
Created and Verified 3D model of industry-level robotic arm and custom end effector using Solidworks
Follow Me!- Autonomous Following Wagon
Excessive hauling of heavy objects often leads to needless injury, and is an aspect of warehouse working that could be alleviated. This project was designed to ease the difficulty of warehouse workers via a trailing wagon. The wagon is designed to hold large loads and maneuver around obstacles using multi-sensor triangulation. We utilize ultrasonic sensors for collision avoidance.
Developed design and technical documentation for autonomous wagon including mechanical design, perception and path planning
Coordinated, determined, and monitored Gantt Charts with Software and Control sub-teams at biweekly sync-up meetings
WordleBot
Wordle is a recent game that has become very popular. In Wordle, a player makes guesses for an unknown word, and at each guess information is revealed about the position of the letters, much like the game Mastermind. The goal is always to guess the word in as little guesses as possible. In this project, we observe 2 different algorithms for solving Wordle, a greedy approach and an expected value gain approach. Our program is able to get a result consistently within 3-4 guesses with both algorithms.
Developed multiple algorithms to effectively solve a Wordle puzzle with feedback given after each round.
Investigated multiple algorithms to contrast their tradeoffs and baseline effectiveness
Captain's Log
"Captain's log, Stardate 43996.2. The Enterprise remains concealed in the dust cloud. And, to my surprise, the Borg have maintained their position waiting for us to come out of hiding." Have you ever watched Star Trek? In Star Trek, all officers keep logs detailing what's been happening while they are on duty. As a kid I found these insanely interesting. These logs were recorded and then could be played back with a transcription of the audio! In this project, we try our hands on transcription via IBM's Watson. The program takes in an audio file and returns a text file with the transcription of the audiofile. Notable departure from Star Trek- the transcription doesn't have any time codes and isn't written in real time, rather the entire file is processed once the file is completed. This is an unfortunate budget limitation of my IBM account. As each portion of the file is parsed, an accuracy score as well as the transcription are returned so that errors can be quickly identified.
Developed algorithms to interpret microphone inputs into arm movement frames using IBM Speech to Text and Python
Integrated transcription software into daily notetaking app for repeated use
DeBlurGAN
As with many image reconstruction tasks, de-blurring images has become leaps and bounds better with the introduction of neural networks. The paper DeblurGAN takes a Generative Adversarial Network to remove blur from a variety of images. In this project, I attempt to recreate the work of DeblurGAN using their established code as a baseline. I retrained the network in the original configuration, as well with several additional Resblocks in order to see if adding complexity improve de-blurring performance empirically. For both iterations I trained the network for 30 epochs using a provided dataset. Adding more ResBlocks does in fact increase clarity of images, leading to less notable artifacts, however due to the additional complexity in the network training time is significantly increased as a result.
Revised publicly available code and datasets for "plug and play" usage
Investigated and improved upon de-blurring research to achieve improved results from original observations
Fido- Animatronic Therapy Dog
Having a pet can often offer a lot of enjoyment to the pet's owner, however as one becomes advanced in years, it can often be difficult to effectively and humanely care for a pet. The goal of this project is to develop an animatronic dog that acts like a normal pet, but requires not care taking such as food or water. We investigate the amount of difficulty it would take to create and verify this type of robot as well as the time it would take to allow it to move in both an effective and realistic way. All models were created in Fusion 360 and simulation was done in Gazebo using ROS.
Designed and modeled quadrupedal robot intended to look and move like a real dog
Simulated similar quadrupedal model in Gazebo and found effective walking gaits and poses to emulate the behavior of a live animal
Performed Finite Element Analysis on simulated model for purposes of safety and load bearing
Contact Me
I am currently open to new opportunities related to engineering, tutoring, or art so please reach out via email or through my other links below!
Resume
EDUCATION
Master of Science in Robotics
Northeastern University
May 2023
Recipient of Dean’s Graduate Scholarship 2021-2023 for outstanding academic performance
Notable Coursework:Advanced Perception, Algorithms,Assistive Robotics, CAD and Manufacturing, Pattern Recognition and Computer Vision, Robotics Sensing and Navigation
Bachelor of Science in Computer Engineering
University of California, Santa Cruz
June 2021
Notable Coursework:Data Structures and Algorithms, Models of Robotic Manipulation
WORK HISTORY
Graduate Student Researcher
Northeastern University
December 2022- August 2023
Formulated, developed, and automated adversarial image testing suite to observe inherent robustness of deep neural networks trained on various image types for use in publication
Conceived and developed metric to normalize change in adversarial image from original in various image types
Investigated the effectiveness of deep neural networks trained on various image types against adversarial images with large amounts of networks and various datasets
Automation Engineer
AcousticaBio
May 2022- August 2022
Developed and implemented 3+ computer vision systems including volume approximation and nozzle alignment using classical computer vision techniques in OpenCV
Synthesized clean documentation for computer vision systems and automation software for use by non-technical employees
Tutor Mentor
University of California, Santa Cruz
September 2020- June 2021
Advised 9 Mentees pertaining to facilitation, time management, and technical material to ensure adequate job performance
Developed 4+ leadership programs with Modified Supplemental Instruction Coordinators at Learning Support Services
Evaluated tutor preparedness and effectiveness of both mentees and other Learning Support Services tutors
Tutor
University of California, Santa Cruz
September 2018- September 2021
Guided students through design and debugging process utilizing various programming languages and design tools
Certified through training in facilitation, leadership, and active learning techniques
Software Engineering Intern
Sequent Software
June 2018- September 2018
Ported and upgraded internal testing suite for public release
Documented and presented new testing features to internal teams and Board of Directors
Quality Assurance Intern
Sequent Software
June 2017- August 2017
Converted internal technical documentation to consumer facing deployment guides
Designed and implemented scripts for unit and load testing
LEADERSHIP
Eagle Scout
Troop 28, Pacific Skyline Council
March 2010- May 2017
Planned, developed, and executed the creation of noise abatement sign in San Carlos airport including seeking funding, receiving city permits, and building signage representing 300+ hours of labor
Recruited and coordinated 15+ volunteers to build infrastructure needed to support noise abatement sign.
New Member Educator
Alpha Phi Omega, Alpha Gamma Nu Chapter
September 2021- April 2022
Pioneered and executed online on-boarding program from scratch in accordance with COVID-19 guidelines
Assembled and Coordinated a team of eight skilled individuals in the execution of on-boarding program
Vice President of Service
Alpha Phi Omega, Alpha Gamma Nu Chapter
January 2019- January 2020
Planned and executed over 50 service events with various organizations, including logistical planning of transport to and from events, as well as briefing and debriefing aforementioned events
Coordinated groups ranging from 5-40+ students to reduce logistical strain from benefiting organizations.