Nested Networks For Robot Control
A. Jansen, Patrick van der Smagt, Groen, F.C.A., Alan Murray · 1994
INTRODUCTION Sensor based robot control systems can overcome many of the difficulties which are caused by unknown or uncertain models of the environment. Conventional sensor based control systems require explicit knowledge of the kinematics and dynamics of the robot arm and a careful calibration of the sensor system. Instead, adaptive neural controllers can be used to build an internal representation of the camera--motor correspondence from exemplar behaviour and adapt where necessary. In that case, static models are not necessary anymore, and the system can cope with changing behaviour of the robot (wear and tear, payload correction) and its sensors (changing lighting conditions, calibration and re-calibration). We concentrate on the pick-and-place task. The system gathers its information from angle sensors mounted on the joints of the robot, and a single camera which is situated in the end-effector of the robot. This placement of the camera increases visual precision when th