Object placement estimation with occlusions and planning of robotic handling strategies

Abstract

This paper describes a system that can find and lift a specific object in a bin containing piled objects. Such a task is ubiquitous in our daily life, for example, in finding a small toy in a toy box or in finding a stationary in a drawer. To efficiently achieve this task, it is necessary to recognize the object placements with consideration of occlusions and to plan a proper hand motions for lifting or searching for the target object. We developed methods for such two necessary functions, with introducing a sweep motion for removing many non-target objects at once. We implemented the methods on a dual-arm humanoid robot with an RGB-D camera and a suction mechanism. The experimental results show the effectiveness of the proposed approach.