Francisco M. López^1,2 (), Theresa Lundbeck¹, Bertram E. Shi³, Jochen Triesch¹; ¹Frankfurt Institute for Advanced Studies, Germany, ²Xidian-FIAS International Joint Research Center, Germany, ³Hong Kong University of Science and Technology, Hong Kong

Fixating objects involves two kinds of eye movements: vergence movements to align the two optical axes on the same point and accommodation control to ensure sharp images. It has long been known that vergence and accommodation control are coupled, but it is presently unclear how they are calibrated during development and how their coupling emerges. Here we propose a computational model based on the Active Efficient Coding framework to explain the joint learning of vergence and accommodation control using intrinsically-motivated reinforcement learning to maximize the mutual information between binocular images and their internal encoding. Training proceeds in a realistic virtual environment with binocular vision and longitudinal chromatic aberration. We compare different model architectures permitting different degrees of interaction between accommodation and vergence control, allowing us to evaluate to what extent these are necessary to replicate experimental results. Our findings suggest that vergence and accommodation can act independently but that early access to visual features used by one controller is enough to replicate the development of a coupling as measured in laboratory experiments.

Acknowledgements: This research was supported in part by “The Adaptive Mind” funded by the Hessian Ministry of Higher Education, Research, Science and the Arts, Germany, and by the Hong Kong Research Grant Council (Grant 16213322). JT was supported by the Johanna Quandt foundation.