Working on developing and automating SAW robots (wave-shaped robots capable of reforming granular terrain).

RC Development Car: Built an RC car and set up the electrical systems, along with a RPLidar and a mini-camera to test our autonomy algorithms and ROS setup. Setup a fully-functional ROS pipeline, with a working autobrake/cruise-control using a new algorithm I designed which accounts for the steering angle of our vehicle. Also setup control systems and localization for the car.Localization: Soldered and setup multiple IMUs for usage in robot-localization, implemented an Extended Kalman Filter to fuse our kinematics and SLAM odometry data.Local Planning in ROS: Made a local planning node based on teb-local-planner, and integrated it with other systems from our vehicle to dynamically plan to avoid obstacles and park in an empty parking spot.Global and Local Planning: Created a custom simulation for occupancy grid-based planning algorithms. Implemented various algorithms like RRT*, LQR, MPC and developed my own which outperformed RRT* by a factor of 8. Connected everything to ROS simulators as well to incorporate into our autonomous vehicle pipeline. Also made a local planner using Casadi, and another using the SciPy optimizer.LiDAR AutoBrake: Created an auto-braking system incorporating Arduino, Raspberry Pi, and an RPLidar with ROS and ROSSerial which automatically brakes a physical autonomous vehicle upon seeing an object within a certain distance threshold. Currently working on developing auto-steer using IMU odometry and local planning as well.

Network Algorithms: Used Network algorithms to develop a new ’Evolutionary Tree’ structure for R&Dexperimental data, created optimal algorithms to use them, and made an API to construct trees from existing data, mine, and visualize them.Regression Analysis: Used data analysis methods, including correlation and regression analysis, to find the most related input and output features in R&D tests. Methods included Phi-K correlation, LASSO, and XGBoost. Also used XGBoost to create models used to predict cost-effective inputs for future tests.Node Tree Visualizer: Created a custom application for visualizing the above tree data using React/ReactFlow with a Python FastAPI backend, using Neo4J and MongoDB as databases.

Worked on replicating Stanford's Universal Manipulation Interface (UMI) for testing in the lab.

Led the programming team of the FRC Wired Boars (7407), helping bring us to semifinals in our division at FRC Worlds in Houston, and winning us the Innovation in Control award. - Developed reliable robot code for the FRC Rapid React and Charged Up games. Incorporated computer vision for ball shooting and tracking, used gyroscopic sensors and fiducial-tracking to keep track of the robots position, and used a variety of other sensors including color, IR, and more to handle the robot's inner mechanisms and game pieces. - Developed a core library for the team to use across separate code-bases for different competitions (7407-Robotpy-Toolkit), and created a GUI to visualize and adjust robot autonomous pathing (AutoBoard). Worked regularly with Python, GitHub, Linux (running on the robot, for vision, and development), and more.

- Taught Elementary school students basic Python skills.- Developed and taught a course in AI with Edge Computers and Tensorflow Lite to High School students

- Created an acoustic gas leak detection system for Boston Dynamics Spot and integrated it with Cognite’s data and robotics software platform. - Used computer vision on the acoustic gas leak detection output to estimate likelihood of gas leaks. Regularly used Python, Linux (on SPOT, servers, and for development), Bash, and Docker. Demoed the product to the client team.

Created a containerized API to allow fast, automatic extraction and transferring of data from scanned medical forms into GCP databases using OCR through OpenCV and the Google Vision API.