# Kunal Kaushik Official portfolio: https://kunalkaushik.vercel.app/ I'm Kunal, an electrical engineering student at the University of Illinois Urbana-Champaign. I build robots, embedded systems, and computer vision tools. Age: 18 Education: B.S. Electrical Engineering, University of Illinois Urbana-Champaign (2026 — 2030) Interests: Embedded, Controls, Sensing, Mechatronics Current book: Steve Jobs — Walter Isaacson Song of the day: Memory Box — Peter Cat Recording Co. Listen: https://open.spotify.com/track/6C5xm2roWdAIda9WJmu1jG Email: kunalkaushik537@gmail.com GitHub: https://github.com/kunalk537 LinkedIn: https://www.linkedin.com/in/kunalkaushik537 Resume: https://kunalkaushik.vercel.app/Kunal%20Kaushik%20-%20Resume.pdf ## Projects ### FSAE Status: Coming soon Category: Robotics More on this work soon. ### NMbL lab with Professor Yim Status: Coming soon Category: Robotics More on this work soon. ### Open-source BLDC motor driver Status: Coming soon Category: Electronics A closer look at the design and development is coming soon. ### Self-balancing humanoid Status: Portfolio project Category: Robotics Exploring two-axis stabilization with reaction wheels, IMU feedback, and custom drivetrains. Dual reaction wheels · ODrive motor control over CAN - Designed perpendicular reaction wheels for stabilization in two axes. - Integrated IMU feedback, Raspberry Pi compute, and a custom 1:8 gearbox. Reaction wheel system 3D-printed with steel dowels around the edges as weight, perpendicular design enables torque in x and y axis; powered by 180KV motors at 24V, controlled by ODrive S1 over CANbus. Raspberry Pi used as microcontroller for complex inverse kinematics and controls equations, connected to IMU data via I2C and processing it + computer vision pipeline onboard. Custom gearbox for driving wheels at 1:8 ratio to maintain stability and torque, utilizing existing motors in slim form-factor through printed modular shell, spur gears, and ball bearings. Stack: Motor Control, Sensor Fusion, Raspberry Pi, CANbus, Python, Soldering, Crimping, 3D Printing, CAD, Inverse Kinematics, Controls Equations Media: - https://kunalkaushik.vercel.app/NovaHumanoidFinal.mp4 - https://kunalkaushik.vercel.app/nova1.png - https://kunalkaushik.vercel.app/nova3.png ### AR ski goggle electronics Status: Portfolio project Category: Electronics Designed a compact ESP32 board to bring motion sensing and an OLED display into ski goggles. 4-layer PCB · IMU + OLED over I2C - Moved from a perfboard prototype to a compact, four-layer board. - Integrated USB-C charging, motion sensing, and a goggle-mounted display. Prototyped on a perfboard with an ESP32 devboard, moved to PCB to minimize form factor and fit in goggles. USB-C connector for convenience and future expansion; serves as power source for built-in 3.7V Li-ion battery charging and serial communication port for flashing. ICM-20948 sensor for IMU feedback with voltage stepped down from 3.3V to 1.8V logic levels; OLED display to create augmented reality effect for information display, both I2C communication 4-layer board (signal, ground, 3.3V, signal) to minimize EMI and maintain impedance, using JLCPCB and component selection through LCSC parts. Stack: Schematic Design, Component Selection, EasyEDA, ESP32, MPU9050, OLED Display, USB-C Connector, Li-ion Battery, TP4056 Charger, 3D-printed Case Media: - https://kunalkaushik.vercel.app/snosight2d.png - https://kunalkaushik.vercel.app/snosight3d.png - https://kunalkaushik.vercel.app/snosight3.png ### FTC competition robot Status: Portfolio project Category: Robotics Designed and built a competition robot with custom mechanisms, odometry, and vision-guided alignment. 5,000+ part assembly · CNC + 3D-printed components - Custom CNC chassis, mecanum drivetrain, and sprung odometry. - Vision-guided intake and multi-stage slides for manipulating game pieces. Chassis: Belt-driven mecanum wheel drivetrain with 4x 12V 435 RPM DC motors at 1:1 ratio, with two-wheel sprung odometry for precise mapping of field movements. Custom CNC-milled aluminum + polycarbonate chassis to balance weight and structural integrity. Intake: Servo-powered linkage driving 3-stage linear slides to intake game pieces at distance. Virtual four-bar geared 2:1 for speedy movement of claw, which has computer vision controlled-yaw to align with rectangular game pieces. Outtake: Four-stage linear slides driven by 2x 12V 312 RPM DC motors with pulley system and kevlar cascade stringing. Second virtual four-bar geared 1:1 with claw to quickly deposit game pieces into basket. Stack: 3D Printing, CNC Milling, Design + Prototyping, Assembly, Soldering, Wiring, Physics + Kinematics, Iteration, Component Selection Media: - https://kunalkaushik.vercel.app/rift1.png - https://kunalkaushik.vercel.app/rift3.mp4 - https://kunalkaushik.vercel.app/rift2.png - https://kunalkaushik.vercel.app/riftresume.png ### Motor driver & sensor board Status: Portfolio project Category: Electronics Built an ESP32 breakout board that brings motor control, servo connections, and motion sensing together. Custom 4-layer PCB · H-bridge + 9-axis IMU - Integrated an H-bridge, servo port, and motion sensing around an ESP32. - Designed power regulation and wireless firmware update support. Powered by a microUSB connector, with 5V battery and LDO to step down to 1.8V for IMU logic levels; ESP32 and servo both at 3.3V. ESP32 serial communication empowered directly through the devboard; initialized with Wi-fi setup to enable wireless flashing and code improvements. H-bridge motor driver for motor control enables basic speed and direction control; used with cheaper motors so IMU serves as encoder replacement to guarantee motion. 4-layer board (signal, ground, 3.3V, signal) to minimize EMI and maintain impedance, using JLCPCB and component selection through LCSC parts. Stack: Embedded, PCB Design, Motor Control, I2C, C/C++, Soldering, 4-layer Media: - https://kunalkaushik.vercel.app/rumble1.png - https://kunalkaushik.vercel.app/rumble2.png ### Wafer alignment detection Status: Portfolio project Category: Computer vision Trained a computer vision model to identify misaligned semiconductor wafers from microscope images. 0.95 F1 score · 1,600-image augmented dataset - Built and augmented a microscope-image dataset, then trained YOLOv9 on an A100 GPU. - Evaluated alignment detection with precision, recall, and F1. Improper wafer alignment during semiconductor manufacturing decreases yield, with current solutions being expensive and physically bulky; CV as a solution. Built dataset using microscope images, annotated with bounding boxes(p: perfectly aligned, n: not aligned) to create 560 image dataset. Heavy class imbalance necessitated augmentation with grayscale, rotation, and color corrections; 1600 images in total. Custom YoloV9 model trained on dataset with 80/10/10 split for training, validation, and test; Training done on Google Colab with 1000 epochs, batch size -1, patience 15, + more; Trained + tested on A100 GPU through Google Colab. Performance metrics calculated with confusion matrix and accuracy, precision, recall, and F1 score. Model achieved F1 of .95, Precision of .962, Recall of .936, overall accuracy of .95. Stack: Dataset Managment, YOLO V9, Computer Vision Pipeline, Python, Google Colab, A100 GPU, Results Analysis, Class Imbalance Handling Media: - https://kunalkaushik.vercel.app/semiconductor3.jpg - https://kunalkaushik.vercel.app/semiconductor1.jpg - https://kunalkaushik.vercel.app/semiconductor2.png - https://kunalkaushik.vercel.app/semiconductor4.png Coming-soon entries are placeholders, not claims of completed work. All metrics are self-reported.