ABSTRACT
In this paper, a systematic connectionist controller design approach is proposed to guarantee stability and desired performance of the robotic system for compliant tasks by effectively combining genetic algorithms(GA) with neural classification and neural learning control techniques. The effectiveness of the approach is shown by using a simple and eficient decimal and binary GA optimization procedures to tune and optimize the performance of a neural classifier and controller, together with tuning of feedback controller.In order to demonstrate the effectiveness of the proposed GA approach, some compliant motion simulation experiments with robotic arm placed in contact with dynamic environment have been performed.
Keywords. Genetic algorithms, Neural networks, Learning Control, Robotics.
ABSTRACT
In this paper, a control scheme based on Neural Networks is described for handling fabrics using robots to feed automated sewing stations. Fabric handling tasks such as sewing demand regulation of the appeared forces and orientation of the fabric. The proposed control approach calculates the appropriate velocity of the robot hand holding the one end of the fabric, while a constant tensional force is applied, when the other end of the fabric is pulled by the sewing machine with an unknown velocity. The effectiveness of the proposed controller lies in the fact that it is independent from the properties of the fabric, as only the measurement of the desired tensional force applied along the fabric is necessary.
Key Words. Fabric handling, Robotic Sewing, Neural Network, Force Control
ABSTRACT
Many vision-based robotic applications either require, or can be improved by, accurate object depth measures. Several computer vision methods exist for extracting depth of features, including stereo vision, structured-light systems, and active monocular depth recovery. Previous efforts using these methods suffered from a variety of problems related to calibration and computational complexity. This paper presents a novel method for active monocular depth recovery that combines new, highly stable active deformable models (snakes) with a structured camera motion along the optical axis to produce depth estimates for all the snake control points.. In experiments with a variety of objects and depths, this method produced control point correspondences and calculated the depth of a large number of control points in the order of 1 ms. Accuracy is demonstrated by results that exhibit errors near the predicted errors when assuming a single pixel mis-measurement in control point location on the image plane.
Keywords. Robotic Grasping, Statistical Dynamic Contours, Eye-in-Hand Robotic Systems.
ABSTRACT
The development of a Fuzzy Neural Network (FNN)-bascd Internal Model Control (IMC) scheme and its application to a dcmotor micromaneuvering system is addressed in this article. The FNN is tuned in an offline manner in order to cancel the motor's inherent nonlinear friction term. The resulting FNN is used in the feedback path, augmented by a primitive linear timeinvariant controller in the forward path. The adjustment of the linear controller's parameters relies on the IMC framework based on the premise that the system's nonlinearities have benn canceled by the FNN. The suggested controller-structure is tested in experimental studies at a dcmotor testbed to investigate its efficiency.
Keywords: Fuzzy neural network, Friction compensation, DCmotor control.