| 講演者 | Professor Dr. Dr. h.c. Gunther Schmidt Department of Electrical Engineering and Information Technology Technische Universitat Munchen, Germany |
|
|---|---|
| 講演題目 | 「Vision-Guided Humanoid Robot Walking: Methodological Approaches and Experimental Results」 |
| 概要 |
|---|
|
Intelligent robot walking needs close interaction between perception and locomotion.This presentation outlines key ideas of a novel approach to perception-based guidance of biped walkers. Starting from a presentation of the developed guidance and control architecture we will discuss in some detail the methodological background of major functional modules such as for step planning and adaptation, for vision for walking and for gaze control. Step planning and adaptation rests upon an approach comprising an off-line and an on-line phase. Based on a definition of a set of step primitives for straight ahead or curve walking, start and stop walking, changing step length or stepping over obstacles appropriate robot joint trajectories are computed off-line by use of mathematical optimization and a high-dimensional kinetic model of the biped. Respective data is stored in a step primitive data base and used in the on-line phase for predictive step sequence planning according to the environmental situation as provided by the visual perception system. Vision for walking comprises a stereo camera pair mounted on a pan/tilt head and a real-time image processing system. Image processing is performed on two time scales: (i) once per step for purposes of scene reconstruction including obstacle classification and localization and (ii) in frame rate for obstacle feature tracking. Respective information forms the base for on-line step adaptation and execution. Gaze or view direction selection and control focuses on two major navigational tasks for biped robot walking: obstacle avoidance and self-localization. Based on the principles of one-step-ahead prediction of visual information and maximization of information content a gaze is selected which optimally serves the two navigational objectives according to the current environmental situation and the walking task. The interaction and the capabilities of the developed functional modules in the framework of the perception-based guidance and control system for intelligent biped robot locomotion in typical indoor walking scenarios is demonstrated through experiments with the 2D-walker BARt-UH and the 3D-humanoid robot Johnnie. |
| 問い合せ先 |
|---|
| 藤田政之
電気電子システム工学科 電話(076)234-4848, FAX(076)234-4870 E-mail: fujita@t.kanazawa-u.ac.jp |