Options
2006
Conference Paper
Titel
Multi-sensor robot control for humanoid two-arm skills
Titel Supplements
Kurzversion
Abstract
Introduction Within the last years a new generation of human interactive robots emerged both in the industrial, public and private environment. Such "humanoid" robots are intended to solve elaborate tasks in close physical contact and cooperation with humans. Due to the complexity of tasks and the uncertainty of a time varying environment the robot has to be equipped with human-like senses which enable it to cope with these extreme conditions. That means that the robot has to be provided with a large variety of sensors and the combination and coordination of the information coming from them has to be one of the main research focuses in order to achieve a completely autonomous humanoid robot. Moreover, in order to deal with different task-classes, the robot has to be able to manage the corresponding specific humanoid skills in a human-like manner. In order to comply with these challenging requirements a multi-sensor Neuro-Fuzzy based supervisory control concept for humanoid robots is being developed by Fraunhofer IITB within the framework of the collaborative research centre SFB 588 "Humanoid Robots" of the Deutsche Forschungsgemeinschaft DFG. Summary Each task of the robot can be decomposed in a sequence of Primitive Skills (PS) which represent basic actions such as position controlled point-to-point motions, force controlled constraint motions, etc. The proposed supervisory control concept represents a hybrid discrete-continuous control system characterized by two hierarchy levels. In the upper hierarchy level a discrete controller reconfigures dynamically the PS sequence choosing the most appropriate one in terms of the identified motion phase or event. In the lower hierarchy level the continuous motion control assures that the currently active PS is managed by the optimal specific controller so that the generated trajectory are accurately followed. Following the proposed concept two implemented skills are presented within this paper. Results The first humanoid skill is focused on the problem of balancing a ball at a defined point of the surface of a tray ("ball on plate" system) by means of the vision feedback given by a stereo camera (eyes). One central problem concerns the uncertain friction model between the ball and the beam. The approach presented in this paper solves the problem by means of a PD control for stabilizing the system in combination with a Fuzzy adaptation in order to optimize its dynamic performances. Moreover an observer in form of a Kalman filter has been implemented to compensate stochastic disturbances and friction uncertainties as well as to balance the time delay associated with the camera measurements. The second humanoid skill performs the friction-based lifting and carrying of heavy and large objects grasped by two arms. Since the grasping is based only on the friction between the grippers and the object, the problem is to control the contact forces providing an optimal adhesion without slip and at the same time without damaging the object due to a too high pressure. Based on a new optical slip sensor a very efficient hybrid force-position control concept has been developed and implemented ensuring the stability of the grasp and providing a better dynamic performance in comparison with the classical master slave approach.