Distributed Decision-making Algorithms for Inspection by Autonomous Robot Collectives
Citation
Ebert, Julia Tenis. 2022. Distributed Decision-making Algorithms for Inspection by Autonomous Robot Collectives. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.Abstract
Inspection is a ubiquitous challenge, from bridges to farm fields to space stations. These tasks are typically dirty, dull, and dangerous, making them ideal candidates for automation. Researchers have already begun to develop algorithms for robotic inspection, but they are typically limited to a few robots performing planned coverage paths with global communication and centralized computation. This creates a single point of failure and scales poorly for larger groups and environments.In contrast, non-inspection research in swarm robotics has developed algorithms for large groups of simple robots with limited sensing and communication, with distributed computation. However, many swarm algorithms solve tasks that share essential features with inspection: robots must (1) move through the environment, (2) sense a feature of their environment, and (3) map those observations to a classification. In this dissertation, I focus on closing the gap between inspection tasks and swarm robotics by developing distributed algorithms to solve two types of inspection tasks: global classification of the state of an environment, and locating faults within an environment.
I present two algorithms that allow a group of simulated Kilobot robots to perform binary classification of a black-and-white world and create a committed collective decision. These algorithms can be conducted without localization or coverage, and with low-bandwidth, small range communication. First, I demonstrate a bio-inspired algorithm built on quorum sensing and honey bee waggle dances, which I also extended with a task-switching strategy to classify multiple color features. Second, I show a Bayesian algorithm to solve the single-feature case, which provides a statistically-grounded strategy that incorporates uncertainty by modeling the world as a distribution.
For robotic target localization, I present a hybrid algorithm built on particle swarm optimization (PSO) to allow simulated robots to locate a value below a threshold in a continuous, monochrome world. I introduce a variable update rate to PSO to improve fault detection, and a dispersion-based movement to share information through the group. The robots are able to achieve a detection success rate comparable to coverage, but without needing to visit the whole environment. I also demonstrate that fault detection with a robot swarm can be applied to the real-world problem of space station fault detection. I employ a related PSO-based algorithm that allows soft-bodied Ferrobot robots to detect multiple vibration sources in a physics-based simulation, and demonstrate that the locomotion and vibration detection can be achieved by real robots in microgravity.
As infrastructure ages and robots become more capable, we can employ collective robotics to ensure safety through inspection. This dissertation demonstrates that we can create robust, interpretable inspection algorithms for large groups of simple robots, without relying on centralized computation or planned coverage. It also shows how a complex task such as inspection can be broken down into fundamental swarm behaviors to make a problem easier to solve; this can serve as an example for using robot swarms to solve other complex real-world tasks.
Terms of Use
This article is made available under the terms and conditions applicable to Other Posted Material, as set forth at http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAACitable link to this page
https://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37372091
Collections
- FAS Theses and Dissertations [6136]
Contact administrator regarding this item (to report mistakes or request changes)