The Spring Sat was my final project for the Classical Dynamics (AA 242A) in my second year at Stanford. I was tasked to design a dynamics problem that involved rigid-body 3d motion and was relevant to aerospace engineering. I chose to create a simplified satellite design and explore the resulting motion if it were hit with a piece of space debris.

I solved the problem using three methods. 1) Conservation of linear and angular momentum; 2) Lagrange's equations; 3) Euler's equations of motion for rigid bodies. Each method returned a set of nonlinear differential equations, which I solved numerically using Matlab. I then created plots showing the satellite's resulting motion. A link to my final report can be found here.

RoboYoga is a yoga instruction application that uses a simulated robot to teach yoga poses to beginners. RoboYoga features a robotic yoga instructor and a robotic avatar that mimics the pose of a human user. The user can select a pose they want to learn from a curated list, and the instructor will demonstrate how to move into the correct position. The user is then prompted to match the pose, and they can see how close they are to the desired position by watching their robotic avatar on the display screen. The motivation for this project was to make yoga instruction fun, interactive, and accessible to all!

A joint space controller sends torques to the instructor robot to achieve the desired pose. A Microsoft Kinect is used to track the joint positions and orientations of the user, and this information is converted into positions and orientations in the simulated yoga studio environment. These joint positions and orientations are then used as inputs into an operational space controller that sends torques to the avatar robot. The final implementation also features a joint space controller for the avatar that sends torques in the nullspace of the operational space controller. This ensures that even if the operational point positions and orientations are properly tracked, the avatar’s joint angles also maintain values that are close to realistic human joint angles.

EyeClue was the culmination of a semester of work for myself and five other Georgia Tech seniors. In our entrepreneurial-focused section of the senior design class, our task was to discover a problem and create a startup company focused on solving that problem. Drawing on the team’s experiences, we found our problem: DUI prosecutions rarely succeed due to the lack of objective evidence gathered during a DUI traffic stop. The current handheld breathalyzers and the long-practiced standardized field sobriety test (SFST) were not producing evidence that would hold up in court.

Our solution was to create a device that would objectively gather data for the most reliable indicator of intoxication in the SFST, the horizontal gaze nystagmus (HGN) test. When the officer holds up their finger and instructs the driver to follow it with their eyes as the finger moves back and forth, the officer is looking for nystagmus, or jumps in the eye’s movement as it scans horizontally. The eyeClue eliminates the officer’s subjectivity in determining the number and magnitude of jumps in the eye’s gaze.

Based on well-documented effects of alcohol on the eye’s gaze, the device employs an acceleration-calculating algorithm to detect jumps and presents the number to the officer. The device also uses an open-source eye-tracking algorithm developed by Google.

My primary responsibility for this project was to develop the jump-detection algorithm. I gathered dozens of videos of sober and drunk subjects tracking a moving stimulus with their eyes, and converted the pupil locations returned by the eye-tracking software into eye angle measurements. This calculation used the subject’s distance from the camera, as well as their interpupillary distance, to convert the pupil position in pixels to the eye’s gaze direction in degrees. From this value, the eye’s angular velocity was found numerically, where spikes would indicate jumps. By setting a velocity threshold for defining the jumps, the device could track the number of times the eye exceeded the target velocity and output the number of jumps to the user.

From this project, I learned the importance of experimental data collection and technical communication. I was the designated Chief Communications Officer for the team and was responsible for writing and editing the majority of our technical reports. I was also responsible for preparing and delivering our final presentation, which won us 1st place in our class of over 20 teams.

The purpose of this research project was to improve the design of a mechanized red onion seedling planter to increase its reliability and feasibility for production on Bangladeshi farms. The original design contained issues that prevented it from functioning correctly with such tasks as feeding the seedlings into the dispenser, correctly orienting them with respect to the ground, and placing dirt around the planted seedlings to keep them from falling over. Along with a graduate researcher, I redesigned the seedling dispenser and burying mechanism. I also designed experiments that were conducted over a three-month period to gather data on the new machine's performance and reliability.

One specific contribution of mine was that I redesigned the burying wheels to include camber in order to more effectively press dirt around the planted seedlings. I created a testing rig using a 3d-printed sleeve that I designed, as well as aluminum inserts to hold the wheel axles that I designed, machined, and threaded myself. This testing rig also allowed for the testing of different wheels placed at different distances apart. I then conducted tests for various wheels and spacings and compared the results to determine the optimal design moving forward.

From this project, I learned the importance of experimental design and of iteration in the mechanical design process. I also improved my report-writing capabilities, as my findings were collected in a single 30-page final report that had to be approved by my research professor.