CARL: RL agent learns to create and steer self-organizing Lenia patterns

danbri · x · 2026-09-19

New paper CARL introduces a closed-loop RL agent that learns to create self-organizing patterns in Lenia, steer their behavior, and let humans guide them in real time. Project website includes interactive demos.

Commentators call it a milestone: RL can now actively control self-organizing patterns toward goals. Combined with NCA+IMGEP and perceptual occlusion as passive controls, the field now has a diverse toolbox for steering self-organizing phenomena.

Original post →

More from Research

Research channel →