Envoy Commander

Collaborative, Centralized, Model-Free Reinforcement Learning

Project Introduction

A brief summary of our project and its vision

Abstract

This project presents the construction and design of a framework for testing and developing a centralized, collaborative, model-free reinforcement simulation. Although similar algorithms exist and the general trend is towards distributed or federated learning, centralized learning can often be cheaper since the field, or “dummy”, agents can be affordable and standardized when controlled by a stronger “commander” machine learning agent that controls the dummy agents when trained or reinforced in some way. This project built this framework and uncovered the significance of sensor noise and slipperiness, even in a controlled environment. Additionally, the framework could potentially be used with many other algorithms as well.

Team

Meet the engineers who brought this vision to life

Yash Gharat

Computer Engineer

Yash Gharat is a senior at the University of Central Florida and will receive his Bachelor’s of Science in Computer Engineering in May of 2022. He plans to continue his education with a Masters in Computer Science while working at CAE Inc. His primary interests are in software engineering and full-stack development.

Anthony Soffian

Electrical Engineer

Anthony Soffian is a senior at the University of Central Florida, and will receive their Electrical Engineering Bachelors Degree in May of 2022. Currently, they plan to find an entry level position in the greater Orlando or Jacksonville area to further expand their experiences as well as their interests in athletics, medical, and education inclined topics on EE or CS pathways.

Andrew Cuevas

Computer Engineer

Andrew Cuevas is a senior at the University of Central Florida and is to receive his Bachelor’s Degree in Computer Engineering in the May of 2022. He has freelance software development experience and intends to look for a more permanent position in the near future, but is open to work in any areas of his interests including creative fields such as dance, music, and game development.

Advisors

Chinwendu Enyioha, PhD

EECS Assistant Professor

Dr. Samuel Richie, PhD

Associate Professor Emeritus

Demos

Documentation

Project Block Diagram

Initial Project Description

Our initial project was constructed with the intention of completing the same task that we ended up including in the final iteration of our project, but utilizing a different approach. We switched from a more maze based implementation to a design that was seamless and more focused on the machine learning aspect of the project, and stayed true to the ideals of scalability, expansibility, and affordability.

The initial design used a game that featured the same three components of the Dummy, the (Envoy) Commander, and the Arena. This arena would later be changed to become the environment but the function of controlling the flow of the scenario is the same.

The initial block diagram focused on the networked association between the dummy agents handling tasks such as color recognition, physical understanding and problem solving, and additionally mechanical navigation using wheels and various sensors. By reducing the complexity of the sensors used and narrowing the tasks each agent in the arena had to complete and understand, the project became more realistic and better adhered to the intentional research and principle centric ideals we had in mind.

For a deeper understanding of the changes that were made, the differences can be seen by exploring the lifetime of the project seen through the documents included within.