抄録
This paper describes our 3D environment model centered intelligent humanoid robot system that the 3D model is used for both motion planning, object recognition and visual modeling. In this approach, we assume that a large majority of objects are known and vision based shape, texture and knowledge modeling are used in order to compliment existing 3D world representation. In the paper, our proposed intelligent humanoid system has capable of developing complex behavior in the daily-life environment since 1) Though our system relay on the model based approach, it does not need to model all objects in the environment, but static objects such as tables and cups. 2) 3D environmental model information is used for directing visual attention for recognizing deformable objects such as strings, fabric and water. 3) In order to model information which is unknown or difficult to pre-programmed, we developed vision based shape and handle information modeling method by manipulating the object by the robot and vision based tool manipulation knowledge modeling method by integratiing of human detection and object recognition.
