Furniture Model Creation Based on Direct Teaching to a Mobile Robot

In this paper, a modeling method to handle furniture is proposed. In real environment, there is a lot of furniture such as drawer, cabinet and so on. If mobile robots can handle these furniture autonomously, it is expected that multiple daily tasks, for example, storing a small object in a drawer, can be performed by the robots. However, it is a perplexing process to give several pieces of knowledge about the furniture to the robots manually. So we propose direct teaching based approach which can easily give not only how to handle the furniture but also an appearance and 3D shape of it simultaneously. The main topic is about the appearance and shape information acquisition using external sensors. I. INTRODUCTION Recently, various robots such as humanoids are actively developed and these robots are expected to work in behalf of human in near future. Because these robots have both human-like many joints and moving ability in real envi- ronment, it is desired to make the robots doing our daily tasks. For example, carry an object and store it in drawers, cabinets and so on. In several researches, robot systems were developed to be assumed to do such tasks in real environment. There were specific requirements as follows: (1) almost knowledge for handling is given manually (4) (5), or (2) ID tags or marks are given to objects (2) (3), or (3) handling procedure is given by tele-operation (6). Our emphasis is that these approaches cause to come on the human when they want to increase handleable objects for the robots. So it is desirable to reduce these manual preparation because there are a lot of objects which can be handled by the robot in real environment. The purpose of this research is to create furniture model which enables robots to handle it. We propose direct teaching based method that human can easily give such model to the robot. Furniture model proposed in this paper is represented as composite model including not only how to handle it but also its appearance and 3D shape. We call this model "IM-model (Instructed Motion Model)". By using this model, it is expected that we can add handleable targets to the robot without specific tools or detail instructions. For creating the IM-model, we need some external sensors which are mounted on a mobile manipulator. In this paper, a single camera and SOKUIKI sensor (7) (compact Laser Range Scanner) are utilized. We propose efficient model creation method by utilizing these sensor and direct teaching. The main topic in this paper is acquiring an appearance and shape information of furniture by using the camera and the SOKUIKI sensor.