About Me

Hello! I'm Sanjay Vijay, an undergraduate at Purdue University studying computer science. I'm passionate about artificial intelligence, computer vision, and robotics. I've also recently become interested in competitive programming. At Purdue, I lead the algorithm subteam of our RoboMaster club, where I've improved computer vision and navigation for our autonomous robots and taught new members the skills to do the same. I'm also a researcher at Purdue's Cognitive Robot Autonomy and Learning (CoRAL) Lab, where I'm working on adapting splatting algorithms for dark environments. This semester, I became a teaching assistant for CS 18200: Foundations of Computer Science and CS 21100: Competitive Programming I. When it comes to industry experience, I've been a summer intern at Draper under the Autonomy and Real-Time Planning Group twice, working on embedded systems, sensor fusion, and vehicle data analysis. Feel free to reach out if you have any questions!

My Resume
User

My Experience

I am a TA for CS 18200: Foundations of Computer Science and CS 21100: Competitive Programming I, supporting over 800 students in total. I hold office hours to assist students with homework and content understanding, using guiding questions and sample problems to enable students to thoroughly learn. I also answer questions on online discussion forums to maximize productive learning. In CS 18200, I collaborate with other TAs to grade homework problems consistently and fairly.

At the Cognitive Robot Autonomy and Learning (CoRAL) Lab, I'm working on 3D mapping from images captured in ultra-low-light environments. I'm exploring splatting techniques like Gaussian Splatting, InstantSplat, and MAST3R. Currently, I'm verifying the results of InstantSplat with PyTorch in our own environment.

RoboMaster is an international robotics competition at the collegiate level. Robots fire projectiles at each other to score points in a match.

Most of my work during freshman year was for automatic aim. I used image segmentation and contour fitting with OpenCV to locate targets on enemy robots. I later did calculations for coordinate transformations from our camera perspective to our cannon perspective. To enhance the aim, I worked on a lookup table to determine cannon pitch from target distance. All of these algorithms run under ROS pipelines that I helped design.

I then became a lead for the subteam. I developed an onboarding assignment for new members and recruited 30 members to the subteam. I also led workshops to teach concepts in computer vision, navigation, and ROS for competition. We also began work on autonomous LiDAR-based navigation for a sentry robot. I used the Nav2 framework to run SLAM on the competition field and tune planners and controllers, eliminating all obstacle collisions.

In 2025, I worked on a library for autonomous underwater vehicle (AUV) analysis. The library's purpose is post-mission data analysis from both simulation and in-water data. The current library used MATLAB and consequently had a license dependency. I transitioned the library from MATLAB to Python, eliminating the licensing cost.

Using C++, I wrote automation scripts to generate data loader Python files for all data objects from logs. I used SciPy and NumPy to load vehicle data from .mat files into Python. With data utilities now in Python, I enhanced the visualizations from the old MATLAB library using the Plotly and Dash frameworks. My new plots were more interactive and formed a more scalable foundation for more data visualization.

I joined the Autonomy and Real-Time Planning Group at Draper, and I worked on two projects in 2024. For my first project, I worked with a team of engineering interns to design a prototype for a lunar micro rover. I led the team's software efforts, starting with the design of a state machine for movement. I later configured a Raspberry Pi 5 to enable I2C communication with our sensors over GPIO. Using Python libraries, I enabled teleoperation of our rover and sensor telemetry. I also made calculations for our drivetrain's inverse kinematics to facilitate teleoperated motion. With my team, I documented design decisions and delivered presentations for design reviews. We ultimately demonstrated teleoperation of our rover in a simulated lunar environment.

My second project involved enhancing an object tracking system using an extension of Kalman filters. The system used Multiple Hypothesis Tracking (MHT), allowing it to consider multiple candidates for an object's position. I thoroughly researched the implementation of the Interacting Multiple Models (IMM) component of our MHT architecture. In a traditional Kalman filter, there is one state transition matrix that serves as a model for expected motion patterns. With IMM, multiple models can have their own state transitions, and the algorithm can output a weighted result based on the likelihood of each model. I implemented IMM with MATLAB and integrated it into an existing Unscented Kalman Filter design, enabling the system to robustly track changing patterns in motion.

I was a student at MIT Beaver Works Summer Institute, a selective summer program organized and hosted by MIT for high school students. I took the Autonomous RACECAR course, where students learn about concepts in robot autonomy and work in groups to program small racecars to traverse race courses. I attended lectures and completed lab assignments to learn about topics like ROS, OpenCV, CNNs, SLAM, Kalman filters, PID control, and LiDAR point cloud processing. Each week ended with a race on a track that tested the content from that week's lectures. Each class also began with a short competitive programming competition.

I led the software implementation for my team's ideas, which involved tuning color segmentation for line following, writing a PID controller from scratch, and implementing wall following with LiDAR data. My team won every weekly race, and for the Grand Prix at the end of the course, we had the best time trial score. Based on the daily programming competitions and an exam on the course material, I was ranked 3rd in my class of 30 students. For my performance in the course, Draper, who sponsored our course, recruited me for a software engineering internship the next summer.

My Projects

I competed in HackTech 2025, a hackathon organized and hosted by Caltech for undergraduates around the world. At the hackathon, I met Ohm Rajpal, Arnab Ghosh, and Vickie Knight, and we collaborated to create Neurosphere, a web application that locates and classifies brain tumors from uploaded MRI scans. It also enhances visualization of the brain, enabling doctors to have a readily accessible map of a tumor's location.

Since I had experience with machine learning, I was responsible for tumor location and classification. I found a dataset on Kaggle with MRI images of brain tumors with labels for tumor type. I used PyTorch to load the pre-trained ResNet CNN model and fine-tune it on the dataset for tumor classification. After achieving 98% test accuracy, I researched methods to locate the detected tumors for improved functionality. I discovered the Grad-CAM method, which uses activation outputs in the neural network to report the most relevant layers. In a CNN, this corresponds to regions in the image where the relevant features are most likely. This enabled our app to highlight regions in the MRI scan where the tumor is most likely located. I implemented this algorithm in our application, representing the output with a heatmap through Matplotlib, and it made a valuable addition to our app. I also used the center and spread of the Grad-CAM distribution to determine the brain lobe containing the tumor and approximate the tumor's size respectively.

My teammates worked on other aspects of the website, including the frontend, backend, database, and the brain visualizer. We seamlessly integrated the model that I had trained, and the result was a fully functional app. We submitted our project to the health track of HackTech 2025, and we won the "Best Use of MongoDB" award. In addition, we were specially chosen for an interview with representatives from Major League Hacking to demonstrate our project for social media.

Neurosphere Screenshot

I worked with Jack Whitman at Blueprint 2024, a hackathon organized and hosted by MIT for high schoolers around the country, to build Safe-Swing, a prototype to prevent dooring accidents. Jack and I are cyclists ourselves, and it's sometimes worrying to ride past a row of parked cars in a busy area. At any moment, you could slam into a suddenly opened car door and face a severe injury. To address this, we wanted to design an effective solution.

Our project involves a detection program connected to an Arduino circuit and a camera stream. The camera stream detects incoming members of traffic and locks the door servo on our Arduino circuit, preventing what would be a dooring accident in real life. We wired an Arduino circuit ourselves and used a Python library to interface with the rest of the program. We used OpenCV to process camera input, and we loaded the YOLOv5 model with PyTorch to detect incoming traffic. We wanted to use the SORT library to add a Kalman filter to track detected objects, but it was difficult to integrate. We developed our own heuristic based on the change in object size to estimate incoming speed.

Blueprint 2024 had a beginner division and a general division, which had over 60 competing teams. Our project won second place in the general division, winning a $300 cash prize and a $1000 scholarship.

Safe Swing Demo Safe Swing Door Safe Swing Award

Contact

Devpost