Swarm intelligence for navigating disordered flows using chemical signals – SWARMER

When a weakly diffusive chemical substance is transported through a turbulent or a porous medium, the resulting concentration field is characterised by a structure made of filaments stretched and folded by the flow. Many organisms have to navigate this complex landscape to find their way to the emitting source, from the bacteria living in our soils, to the cells migrating in our tissues, to the insects pollinating our gardens. These organisms interact with each other and often move together in groups, which allows them to reach the source more efficiently than if they were alone. This project aims at deciphering the mechanisms of this cooperation.

To this end, we will combine computational models of collective motion, direct numerical simulations of scalar transport in turbulent or porous media flows, and optimization techniques from artificial intelligence. Virtual agents will interact within a simulated environment that reproduces the key features of chemical signals found in the wild. To discover the best cooperation strategies, these agents will be trained at navigation using deep reinforcement learning.

Equipped with this unique numerical platform, we will perform in silico experiments to answer three questions:
- Q1: Which individual behaviours allow global navigational abilities to emerge?
- Q2: How does scalar transport in disordered flows constrain navigation and communication strategies?
- Q3: Which features of collective navigation can be conserved across systems and scales?

This mechanistic approach to living systems will uncover the fundamental principles of collective navigation and will contribute to the emerging field of smart active matter. It will also provide guidelines for the design of robotic swarms able to navigate autonomously.