Martin Marietta Corp, Bethesda, Maryland has won a contract from the US Air Force to develop a neural network computer system to enable robots to perform tasks without human intervention. Under the 39-month pact, Martin Marietta Aero & Naval Systems at Baltimore will develop a neural control system for use by the Avionics Laboratory at Wright-Patterson Air Force Base, Ohio. It will be installed on two test beds built by the company – a high performance robotic arm modelled after a human arm and an industrial robot resembling a forklift. The robotic arm will be designed to learn a tough tracking manoeuvre – following with laser vis ion a moving light spot generated by a scanning laser requiring the robot to learn to co-ordinate its laser vision system with smooth, accurate mechanical movements. The industrial robot will learn to use a variety of sensors, including optics and sonar, to determine the location and orientation of objects it will be required to lift and move. The neural net system will be based on a mathematical model of the way humans learn – and forget.
