The Importance Of Humanoid Robots

Humanoid robots must perceive their own state and the state of their environment in order to act successfully. For proprioception, the robots measure the state of their joints using encoders, force sensors, or potentiometers. Important for balance is the estimation of the robot attitude. This is done using accelerometers and gyroscopes. Many humanoid robots also measure ground reaction forces or forces at the hands and fingers. Some humanoid robots are covered with force-sensitive skin. One example for such a robot is CB2, developed at Osaka University. Although some humanoid robots use super-human senses, such as laser range finders or ultrasonic distance sensors, the most important modalities for humanoid robots are vision and audition. Many

Indeed, objects with different 3D shapes often share similar 2D views. Humans are able to resolve this kind of ambiguity easily by producing additional views through object manipulation or self-movement. In both cases the action made provides proprioceptive information that is closely linked to the visual information retrieved from the obtained views. This mode of exploration can be observed already very early in infants, when they start to interact with objects in their environment. Following this principle, an active method has been proposed for a humanoid robot that allows an efficient in-hand object exploration and a perception-driven recognition process. An object is placed in the hand of the robot and during the recognition process it is rotated to produce new views. The object manipulation sequence is not predefined and will be different for every object. For a given, unknown object, it is most efficient to look for a view that yields the most additional information for discriminating it from similar ones. Hence, based on the current view, the associated object probabilities, and the history of actions, a motion is selected that is expected to yield the highest information gain. In our case, actions consist of the rotation of an object in an object centered coordinate system and involve the whole kinematic chain from hand to

In humanoid robots, the sense of touch becomes even more important due to the resemblance of these robots to humans and the fact that touch, is an essential sense for humans. Different routes are being explored in which this new modality can be used in combination with proprioception to identify people’s faces. Although computer vision is now an effective way for robots to recognize a person the process of identification is invisible for the human. It is also very sensitive to illumination conditions and provides only partial information about texture and other individual specific facial features. We seek to provide robots with the ability to create a map of facial features through tactile exploration. This process mimics the way in which blind people build a representation of people’s faces. The task of classifying faces through proprioceptive information, using touch to guide the hand and finger motion across the faces, belongs to the broader range of surface and object recognition techniques. Various approaches have been proposed towards that goal. A series of recognition methods are based on local probing of the object, for instance by creating a tactile image of the contact between a grasped object and the gripper and differentiating between rough flat, edge, cylindrical and