Flow-Lenia: Emergent evolutionary dynamics in mass conservative continuous cellular automata

2025

Read Flow-Lenia: Emergent evolutionary dynamics in mass conservative continuous cellular automata (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-08569.txt.

Introduction

Erwan Plantec and co-authors present Flow-Lenia, a computer model of simple life-like patterns, in a 2025 paper. The model grows out of Lenia, a grid world where smooth rules make moving blobs that look like tiny creatures. Flow-Lenia adds one rule: the total amount of stuff in the world stays fixed, as mass does in real life. The authors tune the rules and watch what forms.

The system makes settled patterns with rich behavior, and search methods can tune it toward forms the team finds interesting. Because each pattern carries its own rule settings inside the flow, many species can share one world and interact. Using activity measures from evolution studies, the authors show early signs of open-ended change, a long-sought trait in artificial life research.

Catalog record

YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-08569

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Flow-Lenia: Emergent evolutionary dynamics in mass conservative continuous cellular automata This manuscript has been accepted for publication in the Artificial Life journal ( https://direct.mit.edu/artl ) ( Plantec et al. 2025a ) Abstract keywords Cellular automata, Artificial evolution, Origins of life, Lenia, Open-ended evolution, Evolutionary activity \auth Erwan Plantec , Gautier Hamon , Mayalen Etcheverry , Bert Wang-Chak Chan , Pierre-Yves Oudeyer , Clément Moulin-Frier \corresponding Erwan [email protected] \affiliations IT University, REAL lab, Copenhagen, Denmark Flowers AI & CogSci Lab, Inria Center of the University of Bordeaux, France Google DeepMind, Tokyo, Japan \abs Central to the artificial life endeavour is the creation of artificial systems spontaneously generating properties found in the living world such as autopoiesis, self-replication, evolution and open-endedness. While numerous models and paradigms have been proposed, cellular automata (CA) have taken a very important place in the field notably as they enable the study of phenomenons like self-reproduction and autopoiesis. Continuous CA like Lenia have been showed to produce life-like patterns reminiscent, on an aesthetic and ontological point of view, of biological organisms we call creatures. We propose in this paper Flow-Lenia , a mass conservative extension of Lenia. We present experiments demonstrating its effectiveness in generating spatially-localized patters (SLPs) with complex behaviors and show that the update rule parameters can be optimized to generate complex creatures showing behaviors of interest. Furthermore, we show that Flow-Lenia allows us to embed the parameters of the model, defining the properties of the emerging patterns, within its own dynamics thus allowing for multispecies simulations. By using the evolutionary activity framework as well as other metrics, we shed light on the emergent evolutionary dynamics taking place in this system. \coverpage Figure 1: We present Flow-Lenia , an extension of the Lenia (a) continuous Cellular Automata (CA). Flow-Lenia (b) introduces a built-in constraint for mass conservation, strongly facilitating the discovery of life-like patterns (c), the optimization of the system parameters towards certain behaviors (d) and the introduction of environmental constraints (e). Moreover, it allows to embed the system parameters within its own local dynamics, leading to large-scale multi-species simulations analysed in the light of the Evolutionary Activity framework (f). (a) Lenia system . The growth U^{t} is computed with kernels K and growth functions G . A small portion of the growth is then added to activations A^{t} to give the next state A^{t+dt} . (section 2 ). (b) Flow-Lenia system . U^{t} is computed as in Lenia and interpreted as an affinity map. The flow F^{t} is given by combining the affinity map and activation gradients. The next state is obtained by “moving” matter in the CA space according to the flow F^{t} using reintegration tracking. (section 3 ). Inset on the right of (a,b) shows 7 patterns obtained with randomly sampled update rules parameters, in Lenia (top, resulting mostly in non-SLP or empty patterns) and Flow-Lenia (bottom, resulting mostly in SLP patterns). (c) Random search . Patterns emerging from random parameter sampling in Flow-Lenia are qualitatively analyzed (sections 4.1 and 5.1 ). (d) Optimizing the system update rule is performed using simple evolutionary strategies with respect to predefined fitness functions, resulting in creatures with specific behaviors (e.g directed motion or chemotaxis). (sections 4.2 and 5.2 ). (e) Environment constraints . Example of environment with food (blue) that creatures can consume to gain mass. (f) Multi species simulations . Snapshots of a large scale multi-species simulation enabled by the parameter embedding mechanism (section 3.1 ), resulting in a evolutionary tree (f.2) (section 5.3 ). 1 Introduction Cellular automata (CA), notably Conway’s Game of Life (GoL) ( Adamatzky 2010a ) , have attracted a lot of interest from the artificial life (ALife) community because of the emergence of life-reminiscent spatially-localized patterns (SLPs). These patterns are of special interest as instances of autopoietic structures (i.e self-produced and self-maintained structures) ( Beer 2004a ) , a fundamental property of life and cognition as proposed in Maturana & Varela 1991a theory ( Maturana & Varela 1991a ) . Recently, Lenia, a generalization of GoL to the continuous domain has been proposed ( Chan 2019a ; Chan 2020a ) . Previous studies with this system have shown the emergence of autopoietic SLPs, also called “creatures”, often resembling microscopic life-forms and displaying various behaviors like motility or self-replication. Moreover, Lenia displays a much wider diversity of emerging patterns than previous systems such as GoL. Such observations have made Lenia a particularly interesting system for studying the emergence of life-like phenomenon and even sensori-motor capabilities ( Hamon et al. 2024a ) . Importantly, these artificial creatures bear a deep similarity to their biological counterparts as enacted agents endowed with constitutive autonomy (i.e creatures showing self-constitution, self-maintenance and some level of behavioral functionality) as described by enactive theories of cognition (see Froese & Ziemke 2009a for a more complete discussion on enactive artificial intelligence). However, those patterns are quite difficult to find, necessitating complex fine-tuned search algorithms in order to find update rule parameters allowing the emergence of interesting patterns (e.g spatially localized patterns). Complex and resilient creatures, displaying features of sensorimotor agency, have been found in Hamon et al. 2024a using intrinsically motivated goal exploration processes (IMGEP) and gradient descent. Another important challenge in artificial life (ALife) and artificial intelligence (AI) is about the design of systems displaying open-ended intrinsic evolution (i.e unbounded growth of complexity through intrinsic evolutionary processes) ( Stanley 2019a ) . Such a process is called intrinsic since no final objective (i.e fixed fitness function) is set by the experimenter, the fitness landscape is intrinsic to the system and depends only on its current state, as in natural evolution where there is no final goal ( Lehman & Stanley 2011a ) . Seminal works by Von Neumann and Ulam in 1951 have paved the way in this direction. They were particularly interested in building an universal self-reproducing cellular automata (CA) capable of achieving open-ended evolution ( Von & Burks 1966a ) quickly followed by Codd’s attempt ( Hutton 2010a ) . Further developments in this direction have quickly followed with Langton’s self-replicating loops, a simpler model of self-replication in CA’s, at the cost of its universality ( Langton 1984a ) . Even though, Langton’s loops were able to self-replicate, no variations could be introduced in the process, thus making the emergence of evolution impossible. Fifteen years after Langton’s self-replicating loops, the goal of obtaining an evolutionary process was achieved by Hiroki Sayama with the Evoloops model which displays Darwinian evolution of self-reproducing Langton’s like loops ( Sayama 1999a ) (see Sayama & Nehaniv 2024a for a more complete account of works on evolution and CA). Emergent evolutionary dynamics have also been studied in the context of neural cellular automata ( Mordvintsev et al. 2020a ; Sinapayen 2023a ) and artificial chemistry ( Kruszewski & Mikolov 2022a ) . However, such systems rely on hand-defined rules, specific structures and controlled settings, ultimately limiting the diversity of patterns that can emerge in the system. On the other hand, even though Lenia creatures display greater diversity, different creatures are governed by different update rules, therefore cannot co-exist in the same world (i.e the same simulation) and cannot interact. Obtaining such an evolutionary process in a CA could be achieved

Read the complete text