Distributed Intelligence with Lighter-Than-Air Vehicles
Abstract
The long term goal of this project is to enable a robust and resilient autonomous system that can support warfighters in dynamic and contested environments with uncertainties. Whether it is decentralized sensing or force multiplication, the deployed system must tolerate noise, disturbances, as well as component failures. This project specifically aims to formalize a method to design robotic agents using sensor-actuator pairs that are distributed in the system.The conventional robot-design approach has two major weaknesses: the centralized architecture reliant on CPU that causes brittleness, and the top-down approach based on idealized mathematical models that forces us to pursue precision. The project will explore the viability of a bottom-up approach in building intelligence intoa robot or a group of robots using individually weak components that interact with each other. The PIs will specifically study how those components can be integrated with analog interfaces for achieving faster response time and better tolerance to imprecision. Through the case-studies using Lighter-Than-Air (LTA) vehicles, the project seeks to characterize the robustness of such systems and construct a generalizable theory for their analysis. The main challenge in achieving these objectives is the lack of systematic design method and the general difficulty in predicting the emergent behavior of a distributed system. In addition, physically-grounded analog components have limited flexibility to function as interfaces when compared to the digital / software approach.To address the above challenges, this project proposes to revive the subsumption architecture which was popular until the 1980s. It proposes to construct distributed systems in layers, starting with the most basic functionality and adding greater and more complex capabilities. Unlike vertical decomposition of the system into sensing, control and actuation (which is now standard), subsumption has robustness built in by ensuring that the lower layers maintain their functionality in the event of malfunctioning at the higher level. For such abottom-up approach, success hinges on the design and implementation of components that are functional in the real world. We will leverage technologies that did not exist 30 years ago and perform rapid prototyping to quickly explore the design space of possible sensing-actuation pairs and analog interfaces.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 24, 2023
- Source ID
- N000142312222
Entities
People
- Daigo Shishika
Organizations
- George Mason University
- Office of Naval Research
- United States Navy