TY - GEN
T1 - Asynchronous and Distributed Multi-agent Systems
T2 - 2nd International Conference on Optimization, Learning Algorithms and Applications, OL2A 2022
AU - Reis, Felipe D.
AU - Nascimento, Tales B.
AU - Marcelino, Carolina G.
AU - Wanner, Elizabeth F.
AU - Borges, Henrique E.
AU - Salcedo-Sanz, Sancho
PY - 2023/1/1
Y1 - 2023/1/1
N2 - Agent-based and individual-based modeling have been widely used to simulate ecological systems. The historical architectures designed to artificial life simulation, namely LIDA and MicroPsy, rely into classical concurrence mechanisms based on threads, shared memory and locks. Although these mechanisms seem to work fine for many multi-agent systems (MAS), notably for those requiring synchronous communication between agents, they present severe restrictions in case of complex asynchronous MAS. In this work, we explore an alternative approach to handle concurrency in distributed asynchronous MAS: the actor model. An actor is a concurrent entity capable of sending, receiving and handling asynchronous messages, and creating new actors. Within this paradigm, there are no shared memory and, hence, no data race conditions. We introduce L2L (a short for: Learn to Live, Live to Learn) architecture, a biological inspired distributed non-deterministic MAS simulation framework, in which the autonomous agents (creatures) are endowed with a functional and minimal nervous system model enabling them to learn from its own experiences and interactions with the two-dimensional world, populated with creatures and nutrients. Both creatures and nutrients are encapsulated in actors. The system as a whole performs as a discrete non-deterministic dynamical system, as well as the creatures themselves. The scalability of this actor-based framework is evaluated showing the system scales up and out − many processes per processor node and in a computer cluster. A second experiment is realized to validate the architecture, consisting of an open-ended foraging simulation with both one or many creatures and hundreds of nutrients. Results from this specific actor-based version are compared to those from a classical concurrency version of the same architecture, showing they are equivalent, despite the fact that the former version scales a lot better. Moreover, results show that exploration of the world is unbiased, leading us to conjecture that our system follows ergodic hypothesis. We argue that the actor-based model proves to be very promising to modeling of asynchronous complex MAS.
AB - Agent-based and individual-based modeling have been widely used to simulate ecological systems. The historical architectures designed to artificial life simulation, namely LIDA and MicroPsy, rely into classical concurrence mechanisms based on threads, shared memory and locks. Although these mechanisms seem to work fine for many multi-agent systems (MAS), notably for those requiring synchronous communication between agents, they present severe restrictions in case of complex asynchronous MAS. In this work, we explore an alternative approach to handle concurrency in distributed asynchronous MAS: the actor model. An actor is a concurrent entity capable of sending, receiving and handling asynchronous messages, and creating new actors. Within this paradigm, there are no shared memory and, hence, no data race conditions. We introduce L2L (a short for: Learn to Live, Live to Learn) architecture, a biological inspired distributed non-deterministic MAS simulation framework, in which the autonomous agents (creatures) are endowed with a functional and minimal nervous system model enabling them to learn from its own experiences and interactions with the two-dimensional world, populated with creatures and nutrients. Both creatures and nutrients are encapsulated in actors. The system as a whole performs as a discrete non-deterministic dynamical system, as well as the creatures themselves. The scalability of this actor-based framework is evaluated showing the system scales up and out − many processes per processor node and in a computer cluster. A second experiment is realized to validate the architecture, consisting of an open-ended foraging simulation with both one or many creatures and hundreds of nutrients. Results from this specific actor-based version are compared to those from a classical concurrency version of the same architecture, showing they are equivalent, despite the fact that the former version scales a lot better. Moreover, results show that exploration of the world is unbiased, leading us to conjecture that our system follows ergodic hypothesis. We argue that the actor-based model proves to be very promising to modeling of asynchronous complex MAS.
KW - Complex agents
KW - Distributed MAS
KW - Situated cognition
UR - http://www.scopus.com/inward/record.url?scp=85148042024&partnerID=8YFLogxK
UR - https://link.springer.com/chapter/10.1007/978-3-031-23236-7_48
U2 - 10.1007/978-3-031-23236-7_48
DO - 10.1007/978-3-031-23236-7_48
M3 - Conference publication
AN - SCOPUS:85148042024
SN - 9783031232350
T3 - Communications in Computer and Information Science
SP - 701
EP - 713
BT - Optimization, Learning Algorithms and Applications - Second International Conference, OL2A 2022, Proceedings
A2 - Pereira, Ana I.
A2 - Košir, Andrej
A2 - Fernandes, Florbela P.
A2 - Pacheco, Maria F.
A2 - Teixeira, João P.
A2 - Lopes, Rui P.
PB - Springer
Y2 - 24 October 2022 through 25 October 2022
ER -