Neurons tuned to binocular disparity in area V1 are hypothesized to be responsible for short latency binocular vergence movements, which align the two eyes on the same object as it moves in depth. Disparity selective neurons in V1 are not only selective to disparity, but also to other visual stimulus dimensions, in particular orientation....[ Read more ]

Neurons tuned to binocular disparity in area V1 are hypothesized to be responsible for short latency binocular vergence movements, which align the two eyes on the same object as it moves in depth. Disparity selective neurons in V1 are not only selective to disparity, but also to other visual stimulus dimensions, in particular orientation.

In this work, we explore the role of neurons tuned to different orientations in binocular vergence control. We train an artificial binocular vision system to execute corrective vergence movements based on the outputs of disparity selective neurons tuned to different orientations and scales. We find that neurons tuned to non-vertical orientations have strong effect on the vergence eye movements. Although adding neurons tuned to non-vertical orientations does not appear to improve vergence tracking accuracy, we find that neurons tuned to non-vertical orientations still play critical roles in binocular vergence control. First, they increase the robustness of policy to the noise. Second, they also increase the effective range of vergence control.

We also compare the effect of different learning algorithms for the control policy. In particular, we compare the use of Attention Gated Reinforcement Learning (AGReL) method, which was used in prior work on this task, with the Natural Actor-Critic method. We show that the performance of the Natural Actor-Critic method surpasses that of AGReL. Although based upon the same input, control policies learned by AGReL can only exploit information about the sign of stereo disparity while the control policies learned by the Natural Actor-Critic method can exploit information about both the sign and magnitude of stereo disparity, leading to faster system responses. Using the Natural Actor-Critic algorithm, we show that the addition of non-vertically oriented disparity selective neurons enables the use of a wider range of vergence commands, which also speeds up the system.

Keywords-vergence control, reinforcement learning, orientation diversity