TY - GEN
T1 - Lunar Site Preparation and Open Pit Resource Extraction Using Neuromorphic Robot Swarms
AU - Thangavelautham, Jekan
N1 - Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2023
Y1 - 2023
N2 - Decentralized multirobot systems offer a promising approach to automate labor-intensive dull and dangerous tasks in remote environments. Use of multiple robots provide inherent advantages including parallelism, fault-tolerance, reliability and scalability in design. However, this is offset by significant challenges from antagonism, when multiple robots trying to perform the same task interfere and undo the work of others causing unreliable performance or worse, gridlock. Conventional approaches rely on domain knowledge of a task at hand. In the absence of domain knowledge, human designers learn the required multirobot coordination strategies through trial and error. Using an artificial Darwinian approach, we automate the controller design process. This approach uses the Artificial Neural Tissues (ANT) that combines a typical feed-forward wired neural-network with a wireless chemical signaling scheme that facilitates self-organized task decomposition. ANT overcomes limitation with conventional and variable topology neural networks to find efficient solutions to a resource-collection task intended for unstructured environments. In this resource collection task, teams of robots need to forage for resources, collect and dump at a designated location. They need to learn to interpret a series of unlabeled cues to identify the dump location and may choose to exploit assets such as a light beacon to home in on these locations. The key to ANT’s advantage is its ability to perform trial and error exploration more efficiently enabling it to acquire creative solutions such as bucket brigades that enable effective cooperation with increased number of robots.
AB - Decentralized multirobot systems offer a promising approach to automate labor-intensive dull and dangerous tasks in remote environments. Use of multiple robots provide inherent advantages including parallelism, fault-tolerance, reliability and scalability in design. However, this is offset by significant challenges from antagonism, when multiple robots trying to perform the same task interfere and undo the work of others causing unreliable performance or worse, gridlock. Conventional approaches rely on domain knowledge of a task at hand. In the absence of domain knowledge, human designers learn the required multirobot coordination strategies through trial and error. Using an artificial Darwinian approach, we automate the controller design process. This approach uses the Artificial Neural Tissues (ANT) that combines a typical feed-forward wired neural-network with a wireless chemical signaling scheme that facilitates self-organized task decomposition. ANT overcomes limitation with conventional and variable topology neural networks to find efficient solutions to a resource-collection task intended for unstructured environments. In this resource collection task, teams of robots need to forage for resources, collect and dump at a designated location. They need to learn to interpret a series of unlabeled cues to identify the dump location and may choose to exploit assets such as a light beacon to home in on these locations. The key to ANT’s advantage is its ability to perform trial and error exploration more efficiently enabling it to acquire creative solutions such as bucket brigades that enable effective cooperation with increased number of robots.
KW - Evolutionary algorithms
KW - Multirobot
KW - Neural-networks
UR - http://www.scopus.com/inward/record.url?scp=85164796512&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85164796512&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-25755-1_25
DO - 10.1007/978-3-031-25755-1_25
M3 - Conference contribution
AN - SCOPUS:85164796512
SN - 9783031257544
T3 - Studies in Computational Intelligence
SP - 365
EP - 384
BT - The Use of Artificial Intelligence for Space Applications - Workshop at the 2022 International Conference on Applied Intelligence and Informatics
A2 - Ieracitano, Cosimo
A2 - Mammone, Nadia
A2 - Di Clemente, Marco
A2 - Mahmud, Mufti
A2 - Furfaro, Roberto
A2 - Morabito, Francesco Carlo
PB - Springer Science and Business Media Deutschland GmbH
T2 - 2nd International Conference on Applied Intelligence and Informatics , AII 2022
Y2 - 1 September 2022 through 3 September 2022
ER -