*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Entraining a Robot to its Environment with an Artificial Circadian System
Matthew O’Brien
Robotics Ph.D. Candidate
School of Electrical and Computer Engineering
Georgia Institute of Technology
Email: mjobrien@gatech.edu
Date: Tuesday, May 4th, 2021
Time: 11:00 AM to 1:00 PM (EST)
Location: *No Physical Location*
Link: https://bluejeans.com/207877095
Committee:
Dr. Ronald Arkin (advisor), School of Interactive Computing, Georgia Institute of Technology
Dr. Ayanna Howard, College of Engineering, Ohio State University
Dr. Céderic Pradalier, School of Interactive Computing, Georgia Institute of Technology
Dr. Magnus Egerstedt, School of Electrical and Computer Engineering, Georgia Institute of Technology
Dr. Zsolt Kira, School of Interactive Computing, Georgia Institute of Technology
Abstract:
As robots become persistent agents in a complex and dynamic world, they must deal with changing environments. Inspired by circadian systems in nature, this dissertation investigates how a robotic agent can both learn the regular cycles and patterns that exist in many environments, and how it can exploit that knowledge.
In this work, relevant environmental states are modeled as time series and forecasted into the future. These forecasts are used to estimate the utility of executing some behavior at any point in time during the dominant environmental cycle. This is incorporated into an action selection system, biasing behaviors to execute at the right times relative to the environment’s dynamics.
As forecasting is inherently unreliable, an error measure is used to adapt the weight of forecasts in action selection, allowing an agent to ignore forecasts when accuracy degrades. As time series models rely on the history for predictions, short disruptions in the environment can degrade forecasting for many cycles. Methods to create modified forecasts which exclude potential outlier data are also presented.
This work was validated using a testbed designed to approximate a precision agricultural task, where a solar powered robot monitors individual plants for pests and weeds. Using this testbed, different environment dynamics and robotic capabilities were tested to explore the scope of the work. Drawing from experimental results, conditions are presented to identify when an ‘artificial circadian system’ will be useful for an autonomous robot.
