Abstract

Evolving visual systems of embodied agents in silico allows us to investigate the mechanisms and drivers of natural vision evolution, test scientific hypotheses, and explore "what-if" scenarios that would otherwise require re-running biological evolution. This provides valuable insights into the development of natural vision. In our framework, embodied agents are placed in single-player games where their survival depends on their ability to evolve their visual systems and learn complex behaviors. This enables an emergence of artificial vision direclty from embodied interactions with the environment.

We find that orientation tasks like navigation in a maze leads to distributed compound-type eyes while an object discrimination task leads to the emergence of high-acuity camera-type eyes. Then we show that how optical innovations like lenses naturally emerge to resolve fundamental tradeoffs between light collection and spatial precision. Lastly, we uncover systematic scaling laws between visual acuity and neural processing, showing how task complexity drives coordinated evolution of sensory and computational capabilities. Our approach shows that embodied agents serve as next-generation hypothesis testing machines while providing a foundation for designing manufacturable bio-inspired vision systems.