No abstract

Deep learning has been widely used for inferring robust grasps. Although human-labeled RGB-D datasets were initially used to learn grasp configurations, preparation of this kind of large dataset is expensive. To address this problem, images were generated by a physical simulator, and a physically inspired model (e.g., a contact model between a suction vacuum cup and object) was used as a grasp quality evaluation metric to annotate the synthesized images. However, this kind of contact model is complicated and requires parameter identification by experiments to ensure real world performance. In addition, previous studies have not considered manipulator reachability such as when a grasp configuration with high grasp quality is unable to reach the target due to collisions or the physical limitations of the robot. In this study, we propose an intuitive geometric analytic-based grasp quality evaluation metric. We further incorporate a reachability evaluation metric. We annotate the pixel-wise grasp quality and reachability by the proposed evaluation metric on synthesized images in a simulator to train an auto-encoder–decoder called suction graspability U-Net++ (SG-U-Net++). Experiment results show that our intuitive grasp quality evaluation metric is competitive with a physically-inspired metric. Learning the reachability helps to reduce motion planning computation time by removing obviously unreachable candidates. The system achieves an overall picking speed of 560 PPH (pieces per hour).

Abstract. Imitation learning is a powerful approach to humanoid behavior generation, however, the most existing methods assume the availability of the information on the internal state of a demonstrator such as joint angles, while humans usually cannot directly access to imitate the observed behavior. This paper presents a method of imitation learning based on visuosomatic mapping from observing the demonstrator's posture to reminding the self posture via mapping from the self motion observation to the self posture for both motion understanding and generation. First, various kinds of posture data of the observer are mapped onto posture space by self organizing mapping (hereafter, SOM), and the trajectories in the posture space are mapped onto a motion segment space by SOM again for data reduction. Second, optical flows caused by the demonstrator's motions or the self motions are mapped onto a flow segment space where parameterized flow data are connected with the corresponding motion segments in the motion segment space. The connection with the self motion is straightforward, and is easily acquired by Hebbian Learning. Then, the connection with the demonstrator's motion is automatic based on the learned connection. Finally, the visuo-somatic mapping is completed when the posture space (the observer: self) and image space (the demonstrator: other) are connected, which means observing the demonstrator's posture associcates the self posture. Experimental results with human motion data are shown and the discussion is given with future issues.

No abstract

No abstract