Vision-based Motion Control of a Biped Robot Using 2 DOF Gaze Control Structure

Abstract

Why biped robot? Although, for this question, we can give various answers, no doubt we are motivated by existence of powerful controller of human brain. Human can exhibit intelligence and a flexible environmental adaptation ability. The concept of embodiment of robot, which is concerned with the human body and the function, allows us to give one of the questions from the point of view of intelligent mechanics. This concept suggests the necessity of the embodiment of humanoid robot in order to obtain an intelligence like human. Recently, in a community of control engineering and control theory area, a control biology and a bio-mimetic control, which aim to elucidate and mimic a control mechanism for various organisms respectively, are interested by many researchers. In their investigations, a humanoid robot, in particular a motion control of a biped robot, has attracted attention for not only similarity of appearances between humanoid robots and humans but also the mechanism or intra-dynamics. For the motion control of biped robots, we can not escape the fact that human motor control depends upon vision. Some recent researches on human motion control (Seara et al., 2002; Seara and Schmidt, 2004) have revealed that vision plays a major role in human posture control systems. Also, for the biped robot, vision is the most important sensory system for his posture and motion control. On the other hand, the humanoid robots nowadays exploit a great variety of sensing technologies such as a angle and torque sensor of each joint, a gyroscope, a pressure sensor in a soul, a CCD camera, a microphone and a range finder. In fact, a humanoid robot shown in Fig. 1, which we use in our experiments, has above sensors in his body. Most of the sensorium of a robot is based on the so-called five senses of human. However, for a real humanoid robot, we have to assert that vision is one of underdeveloped sensors. This is mainly caused by two difficulties: a. Poor Hardware Resources for vision: Hardware performance for robot vision on a biped robot is extremely restrictive due to limitations of the available hardware resources. For a small-sized humanoid robot, we can not use large-sized CCD camera, a heavyweight powerful actuator of the vision system. b. An enormous amount of information in images: For image sequences from CCD camera, we have to extract useful information for robot control by using image processing. How ever, there is no systematic methodology of choosing a kind of visual information from an image.