Discovering and Controlling Diverse Self-Organized Patterns in Cellular Automata Using Autotelic Reinforcement Learning

developmentalsystems.org/rl-for-ca/

Authors: @mcvjetko.bsky.social (attending), @hamongautier.bsky.social and @pyoudeyer.bsky.social

Expedition & Expansion: Leveraging Semantic Representations for Goal-Directed Exploration in Continuous Cellular Automata

developmentalsystems.org/expedition_e...

S. Khajehabdollahi and @mcvjetko.bsky.social (both attending), @hamongautier.bsky.social @ccolas.bsky.social @pyoudeyer.bsky.social

Exploring Flow-Lenia Universes with a Curiosity-driven AI Scientist: Discovering Diverse Ecosystem Dynamics

developmentalsystems.org/Exploring-Fl...

Authors: @thomas-michel.bsky.social and @mcvjetko.bsky.social (both attending), @hamongautier.bsky.social and @pyoudeyer.bsky.social

On the same topic:
How many people are needed for starting a Revolution?
by Mehdi Moussaïd from Max Planck Institute.
The video is in French by auto-translated subtitles work quite well on youtube.

www.youtube.com/watch?v=Y-1q...

Here is the talk outline:

With an interactive website to explore these universes: developmentalsystems.org/Exploring-Fl...

If you want to follow all research from the Flowers group, here is a starter pack:
bsky.app/starter-pack...

For the record, Max Taylor-Davies was able to complete this full project, from idea to paper, in only a 3-month research visit at the Flowers AI and CogSci Lab at Inria. He's nominated for the best student paper award. Max is not yet on Bluesky, here is his X account:
x.com/MaxTaylorDav...

As well as serving as a methodological demonstration for how kin selection and inclusive fitness can be studied using neuroevolution of simulated agents, we believe that our results may have some broader theoretical implications, as explained in this snapshot of the paper discussion:

We show that this is sufficient to evolve altruistic feeding behavior, especially towards kins: hence emergent kin selection

In this new paper, we simply add to this previous model:
- An additional action to feed another agent
- An additional observation indicating if another agent is an offspring
- An infancy period where an agent has reduced abilities

In this paper, we leverage the Non-Episodic Neuro-Evolution framework we developed with @hamongautier.bsky.social and Eleni Nisioti to study eco-evolutionary feedbacks. Illustrated in the fig. below, summarized in this thread: x.com/ClemMoulinFr... , see this paper: dl.acm.org/doi/abs/10.1...