Abstract
The design of complex self-organising systems producing life-like phenomena, such as the open-ended evolution of virtual creatures, is one of the main goals of artificial life. Lenia, a family of cellular automata (CA) generalizing Conway's Game of Life to continuous space, time and states, has attracted a lot of attention because of the wide diversity of self-organizing patterns it can generate. Among those, some spatially localized patterns (SLPs) resemble life-like artificial creatures and display complex behaviors. However, those creatures are found in only a small subspace of the Lenia parameter space and are not trivial to discover, necessitating advanced search algorithms. Furthermore, each of these creatures exist only in worlds governed by specific update rules and thus cannot interact in the same one. This paper proposes as mass-conservative extension of Lenia, called Flow Lenia, that solve both of these issues. We present experiments demonstrating its effectiveness in generating SLPs with complex behaviors and show that the update rule parameters can be optimized to generate SLPs showing behaviors of interest. Finally, we show that Flow Lenia enables the integration of the parameters of the CA update rules within the CA dynamics, making them dynamic and localized, allowing for multi-species simulations, with locally coherent update rules that define properties of the emerging creatures, and that can be mixed with neighbouring rules. We argue that this paves the way for the intrinsic evolution of self-organized artificial life forms within continuous CAs.
Introduction
In the field of Artificial Life (ALife), researchers use Cellular Automata (CA) to study how complex, life-like behaviors emerge from simple rules. Lenia is a modern, continuous version of this, capable of generating mesmerizing "creatures" known as Spatially Localized Patterns (SLPs). However, standard Lenia faces two major hurdles: fragility and global rules. Life-like patterns exist in only a tiny sliver of the parameter space, and update rules are global, meaning different "species" requiring different rules cannot coexist or interact in the same simulation.
Methods
Flow-Lenia addresses these issues by transforming Lenia from a system of spreading activations into a system of moving matter. This is achieved through three key innovations:
- Mass Conservation: Instead of cells simply turning "on" or "off," the system treats the value in each cell as a concentration of matter. The system calculates a Flow ($F$) where matter moves toward areas of high "affinity" but is pushed away from over-concentrated areas to prevent stacking.
- Reintegration Tracking (RT): To move matter across a discrete grid without violating conservation, the authors use RT—a semi-Lagrangian method that "deposits" mass from a source cell into neighboring target cells.
- Parameter Embedding: The parameters defining a creature's behavior (its "genome") are stored in the grid itself. As the matter moves, these parameters move with it, allowing for localized update rules.
Key Results and Insights
Conservation as a Regularizer
Mass conservation inherently prevents the "exploding" patterns that plague standard Lenia. Because the total "ink" in the world is fixed, creatures are naturally constrained to stay spatially localized and are physically robust against perturbations.
Multi-Species Coexistence and Symbiosis
By localizing parameters, Flow-Lenia enables multi-species simulations. The authors observed "coherent creatures" composed of different parameters—for example, one set of parameters forming a protective membrane around a nuclei governed by another set. When different species meet, their parameters can mix via weighted averaging or softmax sampling.
Task Optimization
Using evolutionary strategies (OpenES), the authors successfully trained creatures to perform complex tasks such as directed motion, angular turns, obstacle navigation, and chemotaxis (moving toward a chemical gradient).
Conclusion and Future Work
Flow-Lenia provides a framework for intrinsic evolution where creatures must move toward resources to survive and reproduce. While the paper demonstrates robust multi-species coexistence, future work aims to formalize "agency" and "individuality" within these systems. The ultimate goal remains open-ended evolution, where the system continuously invents entirely new behaviors and biological-like hierarchies.
Summary updated