Impedance control and compliant behavior for safe human-robot physical interaction of industrial robots normally can be achieved by using active compliance control of actuators based on various sensor data. Alternatively, passive devices allow controllable compliance motion but usually are mechanically complex. We present another approach using a novel actuation mechanism based on magneto-rheological fluid (MRF) that incorporates variable stiffness directly into the joints. In this paper, we have investigated and analyzed principle characteristics of MRF actuation mechanism and presented the analytical model. Then we have developed the static and dynamic model based on experimental test results and have discussed three essential modes of motion needed for humanrobot manipulation interactive tasks.

No abstract

Neural Networks are attractive alternative to the classical techniques for identification and control of complex physical systems, because of their ability to learn and approximate functions. This paper presents the development and implementation of adaptive Multilayer Neural Network (MNN) controller in real-time for a drive system. A MNN is first trained off-line to learn (identify) the inverse dynamics of the system, after the training is successfully completed, the MNN is used as a feedforward controller in the control scheme. The advantage of the proposed controller is that the MNN is permanently training. On-line learning is applied while the system is under control to capture any system parameter variations or disturbances. Simulation results are presented to show the advantages of adaptive MNN controller compared to nonadaptive MNN controller. Also, experimental results show that the adaptive MNN controller is able to control the speed trajectory of the chive system with a high degree of accuracy, even in the presence of disturbances.