Neural network control for intelligent end effector of manipulators via proximity sensing

The authors develop a technique using an artificial neural network system for the intelligent control of a robot manipulator (e.g. Unimation PUMA 560) that can grasp objects dexterously with active optical proximity sensors. Multi-layer neural networks are presented to process the sensor information for grasping objects dexterously via iterative learning of the network. The sensory system uses a transmission (beam-block) sensing mode and a reflection sensing mode. Complex nonlinear operations for processing the information can be performed easily by the artificial neural network, and total processing can be performed in real-time by using a DSP board (AT&T WE DSP32C, 49.152 MHz)