Abstract
This research investigated a supervised intelligent control of a robot end-effector using an Artificial Neural Network system. This control system performs dexterous control of end-effector for grasping objects interacting with active optical proximity sensors in real-time. Multi-layer neural networks were presented to process the sensor information for grasping objects dexterously via iterative learning of the network. This intelligent neural network control system was developed as a support system of a vision or voice control system rather than as a stand alone system. The primarily applicable areas of this end-effector control system are rehabilitation for disabled individuals. Other possible applicable areas of this system are the space and the undersea development as well as working in hazardous environments. The intelligent end-effector system utilized a two finger general purpose gripper with optical proximity sensors for sensing the environment. The sensory system used both transmission sensing mode and reflection sensing mode. Complex nonlinear operations for processing the sensor information performed easily by Artificial Neural Networks. The back propagation neural network system was used for the Artificial Neural Network control system. Using a DSP system, real-time control of the intelligent end-effector was accomplished at microprocessor (PC) level. The intelligent end-effector system was developed for use on a Unimation PUMA 560 robot for experimentation.
Kim, Sang-Hee (1992). Neural network control of an intelligent end-effector for rehabilitation robotics. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -1365898.