SENSORY INTEGRATION IN A NEURAL-NETWORK-BASED ROBOT SAFETY SYSTEM

This article presents an architecture for a real-time robot safety sys tem for advanced manufacturing environments. The system is based on ne ural network technology, and contains the neural network detection uni t and the neural network decision unit implemented at the intermediate and high level of processing, respectively. A new, computationally ef ficient methodology for sensory fusion at the intermediate level in th e dynamic robot cell environment is also proposed. In particular, the neural network detection unit is used to combine basic probability mas s functions encoded in certainty grids (local maps) into one final map of the robot environment containing potential collision zones with th e human operator. The map produced by the neural network detection uni t along with other information will be utilized by the neural network decision unit to produce appropriate robot safety decisions. The resul ts of initial computer simulation indicate that the proposed approach can be very useful for design of robot safety in advanced manufacturin g environments. (C) 1995 John Wiley and Sons, Inc.