almARIC - Controlling Growth Instabilities using Reinforcement Learning

• Scientific lead: MATIGNON Laëtitia, PIERRE-LOUIS Olivier (ILM)
• Is SyCoSMA the project lead? Yes
• Start date: 2023-01-01
• End date: 2023-12-31

This project is in collaboration with ILM laboratory, team MMCI.

Morphological instabilities in out-of-equilibrium physical systems are numerous and diverse. One of the most well-known is fingering instability, which leads, for example, to the branched shapes seen in growing snowflakes. This type of instability belongs to a broad class observed across many systems, from hydrodynamics (viscous fingering) to combustion (flame front instability). For several decades, it has been extensively studied, revealing a wide variety of complex patterns that can emerge.

The core idea of our project is to control this instability using physical parameters, allowing us to navigate through the shape space toward a desired morphology. Existing theoretical models are currently unable to predict how to adjust physical parameters to reach a specific target shape. Therefore, we propose to explore model-free control of these morphologies through a reinforcement learning (RL) approach.