Abstract
In this paper, we develop a system that can recognize and grasp objects in complex environment with a dual-arm robot. We use Scale-Invariant Feature Transform for object recognition and pose estimation to cope with a partial occlusion. For grasp planning, we specify in advance pairs of a grasp pose and an approach pose associated with each object, among which we select the one that is at the highest position and verified to be feasible by inverse kinematics and collision detection. To cope with object grasping in various situations, we first classify them into the following three cases: (1) the robot can recognize and grasp the specified object; (2) the robot can recognize the specified object but cannot grasp it due to surrounding objects; and (3) the specified object cannot be detected due to occlusion. The system is implemented as a set of RT-components, which run RT-middleware for their reusability.
