Application of a general learning algorithm to the control of robotic manipulators
International Journal of Robotics Research
Topology-conserving maps for learning visuo-motor-coordination
Neural Networks
INFANT neural controller for adaptive sensory-motor coordination
Neural Networks
Adaptive filter theory (2nd ed.)
Adaptive filter theory (2nd ed.)
Grasping of static and moving objects using a vision-based control approach
IROS '95 Proceedings of the International Conference on Intelligent Robots and Systems-Volume 1 - Volume 1
Neural network real time video processor for early aircraft detection
ETFA'09 Proceedings of the 14th IEEE international conference on Emerging technologies & factory automation
| Hi-index | 0.00 |
A real-time visual servo tracking system for an industrialrobot has been implemented using PSD (Position Sensitive Detector)cameras, neural networks, and an extended trapezoidal motion planningmethod. PSD and directly transduces the light‘s projected position onits sensor plane into an analog current and lends itself to fastreal-time tracking. A neural network, after proper training,transforms the PSD sensor reading into a 3D position of the target,which is then input to an extended trapezoidal motion planningalgorithm. This algorithm implements a continuous motion updatestrategy in response to an ever-changing sensor information from themoving target, while greatly reducing the tracking delay. Thisplanning method is found to be very useful for sensor-based controlsuch as moving target tracking or weld-seam tracking in which therobot needs to change its motion in real time in response to incomingsensor information. Further, for real-time usage of the neural net, anew architecture called LANN (Locally Activated Neural Network) hasbeen developed based on the concept of CMAC input partitioning andlocal learning. Experimental evidence shows that an industrial robotcan smoothly track a moving target of unknown motion with speeds ofup to 1 m/s and with oscillation frequency up to 5 Hz.