Adaptive Robotersteuerung mittels visueller Rueckkopplung (bibtex)
by Thomas Melzer
Abstract:
An adaptive system for kinematic robot control based on visual feedback is presented. The system is capable of moving the effector of an industrial robot to the position of an target object, whose coordinates are extracted from a pair of stationary mounted CCD cameras. The adaptive component consists of an extended neural gas network, which will - without any prior knowledge about camera orientation or robot arm architecture - eventually learn the mapping from stereo image coordinates to associated robot joint angles, the so called hand eye transform. Following the discussion of self organizing systems and the description of the system components, the results of four software experiments are presented, which shall illustrate the impressive performance, but also some weaknesses of the extended neural gas model. Finally, the results of an experiment conducted in a real hardware environment are presented.
Reference:
Adaptive Robotersteuerung mittels visueller Rueckkopplung (Thomas Melzer), Technical report, PRIP, TU Wien, 1997.
Bibtex Entry:
@TechReport{PTR-Melzer97a,
  author =	 "Thomas Melzer",
  institution =	 "PRIP, TU Wien",
  number =	 "PRIP-TR-048",
  title =	 "Adaptive {R}obotersteuerung mittels visueller
                  Rueckkopplung",
  year =	 "1997",
  url =		 "ftp://ftp.prip.tuwien.ac.at/pub/publications/trs/tr-48.ps.gz",
  abstract =	 "An adaptive system for kinematic robot control based
                  on visual feedback is presented. The system is
                  capable of moving the effector of an industrial
                  robot to the position of an target object, whose
                  coordinates are extracted from a pair of stationary
                  mounted CCD cameras. The adaptive component consists
                  of an extended neural gas network, which will -
                  without any prior knowledge about camera orientation
                  or robot arm architecture - eventually learn the
                  mapping from stereo image coordinates to associated
                  robot joint angles, the so called hand eye
                  transform. Following the discussion of self
                  organizing systems and the description of the system
                  components, the results of four software experiments
                  are presented, which shall illustrate the impressive
                  performance, but also some weaknesses of the
                  extended neural gas model. Finally, the results of
                  an experiment conducted in a real hardware
                  environment are presented. ",
}
Powered by bibtexbrowser