Green Thumb: an Autonomous Plant Watering Robot

This project, completed as my senior capstone in robotics alongside two teammates, entailed the ideation, design, construction and validation of a fully autonomous plant watering robot. This robot aims to reduce human labor and effectively maintain plant health by providing high quality care to plants in a home or office setting without the need for complicated irrigation system installations. It contains systems necessary for safely and robustly delivering sustained care including: a 2 degree of freedom manipulator arm, water pump from 9 gallon tank, durable mobile base, high capacity battery, and a suite of autonomy and sensing processes. The autonomy and sensing suite has components such as local and global path planning, SLAM, floor detection, and care scheduling.

Clothing Auto-Modeler

For my capstone in mechanical engineering, as a three person team, we created a device which would take an article of clothing and automatically generate a corresponding 3D model. Presently, there is no commercially available method to 3D scan clothing without capturing the mannequin as well. Moreover there is no device which can scan larger objects without spending upwards of $5000. Our project addresses the gap in existing 3D scanning technology by creating an affordable and effective scanner which could deliver faithful digital renderings of clothing items. The design process we employed included concept generation, analysis of consumer needs, concept selection, engineering analysis, prototype testing, detailed CAD design and validation testing. Our final system was comprised of a precisely controlled stepper motor for rotating the mannequin, custom computer vision algorithms to perform integral image pre-processing, a machine learning based optimizer to determine ideal settings, user-friendly interface, and a robust support structure.

Autonomous Inflation of Elastomeric Ring for Gastrointestinal Deployment

This was a semester long research project focused on creating a contribution to the field of soft robotics. As a team our goal was to design, fabricate and analyze a small elastic ring which could be collapsed into an ingestible pill that would inflate autonomously using a gas-producing chemical reaction once it enters the small intestine. Currently, there are no technolgies which are suitable for long term deployment in the small intestine, nor are there devices which can activate without external equipment, meaning multiple appointments and/or extended medical attention for the patient. My contribution to this project was the dynamic simulation of the device expansion. It provided relevant estimates for the behaviors of the system, including behavior of the uninflated device, optimal device material and geometry, inflation dynamics, and interaction with intestinal lining forces. These simulations were based on meticulously derived analytical equations of motion and integrated using Matlab.

Modeling and Control for Kinematic Chains with Non-Rigid Elements

In robotics right now, the primary way to improve rigidity of robot components is to add material to each linkage, or to shorten said linkage. This leads to a three way tradeoff between total robot workspace, robot mass and reduction in error of robot pose due to material deformation. Through my work in this project, I aimed to deliver a solution to this problem using high fidelity modeling of robot bending and optimal control schemes for the (often) underactuated dynamic systems. Initial work for this goal was completed in the form of controls for the motion of a single beam in 2D, which produced excellent theoretical results. I expanded on this further for a graduate level kinematics, dynamics and controls course in order to model and control an arbitrary 3D kinemanic chain. While my analysis and system simulations shows promising results, computational complexity limited full testing of control schemes. As this projects is the perfect intersection of my interests in mechanical engineering and robotics, I hope to develop my work further until more concrete validation can be performed.

Path Planning with Waypoint Recognition

I strove to expand the functionality of the standard A* path finding algorithm by introducing "waypoints" which corresponded to locations where a set of different paths diverged most - something analagous to a busy intersection. Paths between waypoints then needed to be computed only once and the final graph search could be reduced significantly by essentially skipping over areas where all intermediate paths will have the same or very similar solution. In other words it changes a search of a map of many individual points into a search from start to finish using important "intersections" and using existing solutions to navigate the majority of the map, saving on computation time significantly.

Portfolio Website

I created this website from scratch using HTML, CSS, and JavaScript, so I encourage you to inspect the page!