New Techniques for Inferring L-Systems Using Genetic Algorithm

TL;DR

This paper presents PMIT, a genetic algorithm-based tool that significantly improves the inference of complex L-systems from string sequences, surpassing previous methods in complexity and scope.

Contribution

The paper introduces a novel genetic algorithm approach for inferring complex deterministic L-systems, extending the complexity limit from 20 to 140 symbols.

Findings

PMIT infers L-systems with up to 140 symbols.

It outperforms existing methods in complexity.

Validated on 28 real and artificial models.

Abstract

Lindenmayer systems (L-systems) are a formal grammar system that iteratively rewrites all symbols of a string, in parallel. When visualized with a graphical interpretation, the images have self-similar shapes that appear frequently in nature, and they have been particularly successful as a concise, reusable technique for simulating plants. The L-system inference problem is to find an L-system to simulate a given plant. This is currently done mainly by experts, but this process is limited by the availability of experts, the complexity that may be solved by humans, and time. This paper introduces the Plant Model Inference Tool (PMIT) that infers deterministic context-free L-systems from an initial sequence of strings generated by the system using a genetic algorithm. PMIT is able to infer more complex systems than existing approaches. Indeed, while existing approaches are limited to L-systems with a total sum of 20 combined symbols in the productions, PMIT can infer almost all L-systems tested where the total sum is 140 symbols. This was validated using a test bed of 28 previously developed L-system models, in addition to models created artificially by bootstrapping larger models.