Close

Zach Duguid

Graduate Student

MIT-WHOI Joint Program

View Resume

About Me

I’m a second year graduate student in the MIT-WHOI Joint Program pursuing a degree in Applied Ocean Science and Engineering. In the Deep Submergence Lab at WHOI, we build and test Autonomous Underwater Vehicles (AUVs). AUVs can be used for a variety of tasks in oceanography, including: monitoring the health of coral reefs, searching for new hydrothermal vents, and studying Arctic sea-ice. My research focuses on real-time sonar processing to improve navigation and sensing onboard the AUV, which is important for vehicle survivability and accuracy of data collection. The photo on the left shows me collecting sea-ice measurements on the Saint Lawrence River in Saint-Fabien, Canada. Outside of research, I enjoy staying active by skiing, cycling, and running.


Education

MIT-WHOI Joint Program

Aug 2018 - Present

Master of Science in Mechanical Engineering
Adviser: Richard Camilli

Massachusetts Institute of Technology

Aug 2014 - Jun 2018

Bachelor of Science in Aerospace Engineering
Minor in Computer Science


Experience

Australian Centre for Field Robotics

Visiting Researcher

With ACFR, I implemented a Generative Adversarial Network (GAN) machine learning architecture to make bathymetry predictions given sparse sonar readings, a prediction problem similar to image inpainting. To improve learning, I generated large sets of training data by simulating vehicle dynamics and sonar measurements.

Computer Science and Artificial Intelligence Laboratory

Undergraduate Researcher

With CSAIL, I helped deploy an array of AUVs off the coast of the Hawaiian Islands to demonstrate human-robot interaction, multi-agent execution, and adaptive sampling techniques in a challenging ocean environment. Contributing to these efforts, I developed energy-aware path planning capabilities using a chance-constrained MDP framework.

Woods Hole Oceanographnic Institution

Summer Research Fellow

With WHOI, I created a graphical user interface to monitor the battery state of the Slocum Glider vehicle. To take advantage of the long-endurance capability of the Slocum Gliders, I performed vehicle range analysis for different power mode scenarios and ocean current conditions. From the hardware perspective, I also designed and built the internal battery pack chassis to increase strength and decrease weight for the glider.

Northrop Grumman

Systems Integration, Test, and Evaluation Engineer

With Northrop Grumman, I programmed a Google Earth visualization tool that displays flight data from the Global Hawkaircraft by assimilating and synchronizing state variables across multiple data files. Along with developing a flight visualization tool, I also operated software and hardware components of the Global Hawk in order to conduct systemand subsystem level testing for segment integration.

Man Vehicle Laboratory

Undergraduate Researcher

With MVL, I assessed the accuracy of the Enhanced Dynamic Load Sensor for the International Space Station (EDLS-ISS), a platform used for strength training in microgravity environments. To do so, I extracted motion data from test subjects performing various weightlifting movements while undergoing microgravity via NASA’s parabolic flight program to develop a musculoskeletal model.

Hong Kong University of Science and Technology

Undergraduate Researcher

With HKUST, I analyzed protein localization in yeast cells to identify novel protein pathways in retrograde transport. While doing so, I learned common wet lab operations such as cell transformations, DNA extractions, and PCR amplifications.


Projects

Perception, Planning, and Control for Autonomous Racecar

The autonomous car challenge had two core requirements: the ability perceive cones of different color while avoiding collisions with said cones, and the ability to race around a track without colliding with walls or spinning out. This challenge tied together three core principles of autonomous systems: perception, planning, and control. Two stereocameras and one Lidar scanner were used to sense the environment, and ROS was used when implementing the software systems.

View Project

Adversarial Autoencoder for Predicting Bathymetry Given Sparse Sonar Data

Autonomous Underwater Vehicles (AUVs) are commonly used for ocean exploration, often in dangerous and unknown environments. Due to the power constraints and risks involved, AUVs are not typically equipped with expensive high-resolution sensors. As a result, instruments such as acoustic sonar only achieve a partial coverage of the underlying bathymetry. To maximize the information gain from partial coverage sonar measurements, I propose a GAN-based approach to predict the missing sonar information. Specifically, I propose an Adversarial Autoencoder (AAE) structure that utilizes adversarial training to perform inference.

View Project

Battery Chassis Design for Slocum Glider

The Slocum Glider AUV contains 3200 Whrs worth of charge spread across four unique battery packs, which enables the vehicle to conduct long-range survey missions. In order to reach this power capacity, I carefully designed the battery chassis for each individual battery pack. I utilized computer aided design (CAD) software to iterate on the design of the chassis. Design decisions were made to maximize strength, minimize weight, and minimize the volume footprint within the hull of the glider. 3D printing construction methods allowed me pursue highly customized designs that would be been nearly impossible to construct in a machine shop.

Energy-Optimal Path Planning for AUVs

From vehicle range analysis that I conducted at WHOI, it is shown that large performance gains can be achieved by optimizing AUV speed subject to both hotel load scenarios and ocean current conditions. In this context, hotel load refers to the power load of all non-propulsive systems. To demonstrate the advantage of varying AUV speed, I developed a speed controller that calculates the optimal speed given the vehicle's thruster efficiency along with the hotel load scenario and ocean current conditions. This controller was implemented to support a field robotics campaign off the coast of Hawaii for the Slocum Glider vehicle.

View Project

Optimal Rapid Exploring Random Trees

For some path planning scenarios, incorporating random sampling methods can be highly beneficial. Rapid exploring Random Trees (RRT) randomly samples from high dimensional configuration spaces to discover branching paths which quickly lead to a solution of the path planning problem. One issue with the general RRT algorithm is that is proven to yield sub-optimal results. In this project, I implement an optimized version of the RRT algorithm that updates the branches of the tree as exploring progresses in order to promote optimality in the solution.


Technical Skills


Interpersonal Skills

Get in Touch