Joint learning for smooth pursuit eye movement and moton parallax through active efficient coding

Estimating depth by using vision system has been continuously researched for a long time. There are a lot of works that can estimate depth by using binocular disparity. However, there is little work on depth estimation by using monocular depth cue. Some of them require specific condition such as environment, some requires calibration. So, if there are some changes or interferences in environment or configuration of vision system, the solution seems to fail later. In order to overcome this problem, our goal is to propose an autonomous learning and self-calibrating for active depth perception system based on motion parallax by mimicking human being which is robust and able to adapt to novel environments without prior knowledge. In this research we propose a hardware-independent learning model can autonomously understand motion parallax according to different depths by using joint development of visual encoding and smooth pursuit eye movement This joint development is achieved by design of general cost function for reinforcement learner. Finally, we verified the proposed model by using simulator called V-REP.