We have proposedmotion sketch [1] as a method to represent an interaction between a one-eyed learning agent (a mobile robot) and its environment. In this paper, we extend the basic idea inmotion sketch, tight coupling of perception and action, tostereo sketchby which a stereo-vision based mobile robot learns various kinds of behaviors such as target reaching and obstacle avoidance. Here, we deal with target reaching behavior with whichmotion sketchcopes by detecting and avoiding occlusions. First, an input scene is segmented into homogeneous regions by the enhanced ISODATA algorithm with MDL principle in terms of image coordinates and disparity information obtained from the fast stereo matcher based on the coarse-to-fine control method. Then, the segmented regions including the target area and their occlusion status identified during the stereo and motion disparity estimation process construct a state space for a reinforcement learning method to obtain a target reaching behavior. As a result of learning, the robot can avoid obstacles without describing them explicitly. We give the computer simulation results and the real robot implementation to show the validity ofstereo sketch.