Michael Louis Iuzzolino

Adventures in Deep Learning, Robotics, and Biology

Current Research

Deep Learning Robotics / Transfer Learning

top

Overview

The problem of robot autonomy in the navigation of outdoor man-made trails is challenging and mostly unsolved. Navigating these environments is important for applications that include: wilderness mapping, search and rescue, and firefighting. Previous research has seen success in robot trail navigation using neural networks and large image datasets of real outdoor trails. However, we propose a solution that bypasses the need to collect real images of the outdoors and instead utilize synthetic data gathered from virtual environments. Using a recurrent neural network architecture trained on synthetically generated trail data we achieve classification accuracies greater than 58% on test datasets of previously unseen real life trails, demonstrating the feasibility of virtual to reality transfer learning.

Demo

Paper

In development.

Interactive Machine Learning for Collaborative Human-Machine Perception

top

Overview

This project aims to explore interaction architectures between human military analysts and machine learning systems in the domain of intercontinental ballistic missile (ICBM) detection and tracking. More information coming soon.

Demo


Human-Robot Interactions

top

Overview

In this study we are investigating the effects of form and collaboration on the positive treatment of robots. Experiment In Progress. More information coming soon.


NASA SUITS - Augmented Reality and Spacesuit Informatics

top

Overview

The NASA Spacesuit User Interface Technologies for Students (NASA S.U.I.T.S.) Design Challenge is a mission-driven project in which university student teams design and create spacesuit informatics using an augmented reality (AR) Microsoft HoloLens platform. Informatics are the avionics of a spacesuit which help an astronaut become more efficient and effective during a spacewalk, often in the form of visual displays. The student-designed visual display and audio environments will present information to aid astronaut subjects in performing simulated Extravehicular Activity (EVA) tasks. After developing their environment, selected student teams will have the opportunity to travel to NASA Johnson Space Center to test their prototypes in an on-site facility.


Deep Learning Biology

top

Overview

Deep learning tools will be applied to pattern recognition problems in biology, such as bacterial persistence and protein sequence prediction and Malaria network analysis. More information coming soon.



Previous Research

FlapPyBI:O - Neuroevolution of Augmenting Topologies

top

Overview

This project was carried out over the Fall 2016 semester as part of a graduate course in Machine Learning at CU Boulder. Our team developed an algorithm capable of learning to play the game FlappyBird. We utilized Neuroevolution of Augmenting Topologies (NEAT), a genetic algorithm that generates neural networks of increasing complexity. The evolutionary process was driven by a fitness function contrived to reward neural networks capable of traveling the furthest distance, as well as minimizing energy expenditure (measured as number of 'flaps'). Our project and the methods are summarized in the video below, which has been utilized by many individuals to implement their own versions of NEAT on a variety of unsupervised learning problems.

Demo

Paper

This browser does not support PDFs. Please download the PDF to view it: Download PDF.</p> </embed>

Space Habitat Design

top

Overview

This project was carried out over the Fall 2016 semester as a part of a graduate course in Space Habitat Design at CU Boulder. Our team developed a space habitat geared toward space tourism; a space HOSTEL (Habitable Observatory for Science, Tourism, and Expeditionary Living) suitable for LEO / HEO operations. I served as the project team lead, primary lead on crew accommodations, secondary on systems engineering, as well as having developed all 3D models (completed in Blender) and videos of mock operations.

Demo

Paper

This browser does not support PDFs. Please download the PDF to view it: Download PDF.</p> </embed>