About
I am a Robotics PhD student interested in the intersection of dynamics, motion planning, and machine learning. I currently work in DART Lab with Dr. Anirban Mazumdar and I am expecting to defend in Summer 2023. I also work as a year round graduate intern for Sandia National Laboratories. My work explores the application of deep reinforcement learning to learn effective motion primitives for fixed-wing aircraft to improve performance of symbolic planning algorithms. I completed a Bachelor’s in Mechanical Engineering and Minor in Computer Science at Georgia Tech in 2018, during which I also worked in CRAB Lab with Prof. Daniel Goldman to prototype a snake-like robot and gantry system.
Experience
Graduate R&D Intern
Sandia National Laboratories
May 2022 – Present
Graduate Research Assistant
DART Laboratory – Georgia Institute of Technology
August 2018 – Present
Publications
Utilizing Reinforcement Learning to Continuously Improve a Primitive-Based Motion Planner
Zachary Goddard, Kenneth Wardlaw, Kyle Williams, Julie Parish, and Anirban Mazumdar
Motion primitives provide a fast and powerful means of solving difficult kinodynamic planning problems. This work presents and demonstrates a framework for learning motion primitives that improve the performance of a Hybrid A* planner for navigation with fixed-wing aircraft.
High Force Density Gripping with UV Activation and Sacrificial Adhesion
Esther Lee, Zachary Goddard, Joshua Ngotiacco, Noe Monterrosa, and Anirban Mazumdar
High force density and low power consumption are critical for mobile systems due to limitations on payload and energy storage. This work explores the use of UV cured adhesives to grip surfaces in unstructured environments. This design could be particularly useful to long duration perching in environments without designated attachment points.
Adversarial Sampling-Based Motion Planning
Hayden Nichols, Mark Jimenez, Zachary Goddard, Michael Sparapany, Byron Boots, and Anirban Mazumdar
Deceptive path planning poses the problem of finding a path to a goal while preventing an observer from identifying the desired destination. This work presents Adversarial RRT* which extends the RRT* algorithm with a learned cost function to balance the cost to reach a goal and an observer’s accuracy in selecting the intended goal.
Dynamics of Scattering in Undulatory Active Collisions
Jennifer Rieser, Perrin Schiebel, Arman Pazoiki, Feifei Qian, Zachary Goddard, Kurt Wiesenfeld, Andrew Zangwill, Dan Negrut, and Daniel Goldman
Self-propelled systems experience complex interactions with their surroundings which may be not be characterizable by collisions studied in classical physics. This work utilizes a robophysical model to experimentally study the effects of such interactions on an undulatory, self-propelled system.
Projects
Research-Related
JSBSim Gym Environment
Github: https://github.com/zgoddard3/jsbsim-gym
This project demonstrates an environment similar to the one used in this paper. The environment simulates an F-16 flight dynamics model using JSBSim and provides a goal for the agent to attempt to reach. The repository includes scripts for training and evaluating a Soft Actor-Critic (SAC) model on the environment using Stable Baselines 3. I also provide a custom feature extractor that transforms the observation space to values I found to be most beneficial for this task. Further details can be found in the code comments in jsbsim_gym.py and features.py. The animation below demonstrates a successful simulation using the trained SAC agent.
Coursework
Adaptive Control
Github: https://github.com/zgoddard3/adaptive-control
In this course project I designed several controllers for a planar model of a ducted fan. The controllers include a linear baseline, MIMO DMRAC, and DMRAC with non-linearities in the span of the control.
Extracurriculars
Game Development
During my time at Georgia Tech I have collaborated on 8 different video game projects with the VGDev Club. I’ve highlighted a few of them below, but all my forks can also be found on my Github.
Path to Harmony
VGDev Link: https://vgdev.gtorg.gatech.edu/game/path-to-harmony/
Github: https://github.com/zgoddard3/PathToHarmony
Path to Harmony is a turn-based fantasy tactics game. During this project I worked on the enemy decision making for selecting targets based on unit types (e.g. archer, foot soldier, cavalry, etc) and terrain.
Hengliding
VGDev Link: https://vgdev.gtorg.gatech.edu/game/hengliding/
Github: https://github.com/zgoddard3/Hengliding
Hengliding is a chicken raising and racing game. During this project I designed the aerodynamic model for the racing portion as well as the steering AI for the other racers.
Lunacia
VGDev Link: https://vgdev.gtorg.gatech.edu/game/lunacia/
Github: https://github.com/zgoddard3/Lunacia
Lunacia is a 2D puzzle adventure game. During this project I implemented various game mechanics and interactions. I also helped design levels, puzzles, and a few visual effects.