Mechanical and Aerospace Engineering

MAE Ph. D. Exit Seminar- By David Ko

A Persistent Decentralized Massively Scalable Mobile-Agent
Based Evolutionary Algorithm for Optimizing Robotic Gaits

By: David Ko
Advisors: Professor Harry Cheng

Multi-robot and modular robotic systems present a novel robotics paradigm: They can
accomplish a wide variety of tasks using different physical configurations and control laws; they can
be cheaper than traditional robotics because identical modules can be mass produced; they form
fault-tolerant systems because malfunctioning modules can easily be swapped out; and their
available computing power continues to increase exponentially. These factors can make autonomous
modular robots ideal for usage in remote areas where a single set of robots might have to perform a
variety of impromptu tasks. For instance, these robotic systems might be used for search-and-rescue,
or deep-sea and space exploration missions. Creating control laws for a cluster of modular robots
with an arbitrary morphology presents an interesting challenge. Ideally, the robot morphologies and
control laws for arbitrary morphologies should be generated dynamically by the robotic systems
themselves for cases where human input is minimal or nonexistent. Among these controls is the
locomotion of these modular robots, or robotic gaits.
A Persistent Distributed Agent-Based Genetic Algorithm (PDABGA) has been designed and
tested towards the goal of dynamically generating robotic gaits for a cluster of modular robots in the
course of this research. PDABGA is based on an agency framework called Mobile-C, an IEEE
Foundation for Intelligent Physical Agents standards compliant mobile agent framework. PDABGA
leverages innate features of the mobile agent framework, such as persistence, computational
distribution, and robustness. To generate optimal gaits for arbitrary robot morphologies, PDABGA
augments agents with the core principles of the Genetic Algorithm: sexual reproduction among a
population of candidate solutions. The ivsystem has been validated to show that it can leverage
multiple cooperating computing nodes to generate high quality results.

Date(s) - 09/25/2014
2:00 pm - 4:00 pm

Bainer 2033


Loading Map....