Future robotic systems intended to be used for domestic housekeeping or elderly care should flexibly adapt to the environment which is uncertain and may fluctuate. However, the current robot systems for industrial domain can operate only in the limited environment and require the special knowledge of the robot motion instructors such as how to use the teaching-pendant and program robot motions. Obviously such requirements are not appropriate for home-use robots. This paper proposes a robot sequence generation method based on the robotizing strategy considering required adaptability to environmental fluctuations and task difficulty for humans and robots. With the proposed method, domestic housekeeping tasks can be classified into three types such as automated tasks with hand-guiding instruction when necessary, human-robot cooperation tasks, and manual tasks without using robot. Then, it is shown that robot motion sequence can be described by a unified framework. The proposed framework is an extension from the one for in-situ robot motion modification method by hand-guiding instruction which was already proposed by the authors. The proposed method is actually applied to a sequence of cooking tasks and experiments are conducted by using a lightweight pneumatic robot arm. The experimental results show the effectiveness of the proposed method.

抄録全体を表示

Robots which play important roles such as social welfare and services in the home and office are required for the aging society. These robots need a location system which can be adapted in the home environment. Therefore, in this paper, we develop a location system based on comparing of 3D surrounding model with input images of a camera. In this system, corner features such as corners of doors or furnitures, are extracted from inputted images of a camera by ORB. A position and a posture of the camera are estimated by matching these corner features and those of 3D surrounding model. Then, the estimation of the position and the posture is continued by tracing these features using FREAK. The location system can estimate the position and the posture of the camera in a case that the camera is rotated on the optical axis. Furthermore, a mobile robot system which is used the proposed location system is constructed. Using the mobile robot system, traveling experiments are performed. As a result, the mobile robot system could follow the reference path based on the estimated position and posture of the camera, when the camera was rotated on the optical axis.Thus, an effectiveness of the developed location system is demonstrated.

抄録全体を表示

In a robotic cell, assembly robots have to grasp parts in various shapes of a target product accurately and robustly against some external wrenches (i.e., forces and moments) exerted on the parts in order to ensure assembly tasks. Assuming the use of a universal robotic hand with multiple fingers for this purpose, it is desirable that necessary finger forces are as small as possible. From this point of view, in this paper, we consider grasping of three-dimensional parts by a universal robotic hand with parallel stick fingers, and propose a method to optimize the grasping under a given required external wrench set. The utility of the proposed method is shown with numerical examples.

抄録全体を表示

This paper describes a method of unfolding an item of clothes by a single arm robot. The unfolding motion is assumed to use a corner of a table so that a crumpled cloth is unfolded by pulled up while making contacts with the corner. Using this approach, two problems arose; entrainment problem and outward warp problem. To solve the former, we proposed an instrument to prevent entrainment. To solve the latter, we built an algorithm that recognizes a warping form of clothes by SVM. Using four kinds of clothes, we confirmed the effectiveness of the proposed method by using a real single arm robot. Our experiment resulted at a success rate from 65 percent to 85 percent.

抄録全体を表示

This paper is aimed at reducing the amount of knowledge to avoid lower learning performance of an agent in transfer learning. In transfer or multitask reinforcement learning problems, the agent reuses policies which were learned in past tasks in order to efficiently solve unknown tasks. Therefore,the agent has a large number of state-action pairs as knowledge. But, at the same time, it causes both explosively increasing the amount of knowledge and decreasing the learning speed. This paper proposes a method for reducing the amount of knowledge on the basis of value. The effectiveness of the proposed method was verified with the simulation of the reaching problem for a multi-link robot arm. The proposed method achieves a reduction of the amount of knowledge and learning time. It also improves learning performance of the agent.

抄録全体を表示