Jure Čas

Uncalibrated visual servo control with neural network

Research into robotics visual servo systems is an important content in the robotics field. This paper describes a control approach for a robotics manipulator. In this paper, a multilayer feedforward network is applied to a robot visual servo control problem. The model uses a new neural network architecture and a new algorithm for modifying neural connection strength. No a-prior knowledge is required of robot kinematics and camera calibration. The network is trained using an end- effector position. After training, performance is measured by having the network generate joint-angles for arbitrary end effector trajectories. A 2-degrees-of-freedom (DOF) parallel manipulator was used for the study. It was discovered that neural networks provide a simple and effective way of controlling robotic tasks. This paper explores the application of a neural network for approximating nonlinear transformation relating to the robots tip-position, from the image coordinates to its joint coordinates. Real experimental examples are given to illustrate the significance of this method. Experimental results are compared with a similar method called the Broyden method, for uncalibrated visual servo- control.

Neural network position control of XY piezo actuator stage by visual feedback

This paper describes the visual feedback positional control of the XY piezo actuator stage (PAS). The XY PAS control system consists of four main components, i.e. XY PAS as controlled object, supply electronics for piezoelectric actuators (PEAs), microscope with digital camera for visualization and for measuring the actual position and a vision processing module in combination with a desktop PC, as processing hardware. XY PAS is fabricated by a photo structuring process from photosensitive glass, and PEAs are built-onto meet the request for its precise movement. It is evident from the electromechanical model of XY PAS, that accurate positioning of XY PAS is an exacting piece of work, due to the nonlinear hysteresis inherent in PEAs. Accordingly, two neural network control techniques were developed, i.e. the feedforward neural network controller (FFNNC) and the feedforward/feedback neural network controller (FF/FBNNC). Proposed neural network controllers are compared with the traditional linear controllers.

Micro and nano robotics

The paper presents two techniques for controlling of micro/nano robots. The first one uses so called visual servoing techniques, where the robot tip position is measured with a camera. The position of the robot tip is extracted from the live video picture, so on-line position feedback control can be established. The experimental results of 2 DOF micro robot neural network controller with resolution of 500 nm is presented. The second presented experiment will show research results of 5 DOF nano robot cell with a gripper with resolution of 62 nm. The so called adaptive bang-bang position controller will be presented. The position of each linear axis is measured by a magnetic incremental encoder. The presentation focuses to the experimental results achieved by so called interface for prevention of collision in the nano world. We use a haptic device to get filling of a touch in the nano world.

A particle swarm and neural network approach for position control of XY Piezo Actuator Stage

This paper describes a position control of 2 degrees of freedom (DOF) XY Piezo Actuator Stage (XY PAS) with Feedforward Neural Network (FNN) and additional Particle Swarm Optimization (PSO) approach, which is used as an improved learning method for optimizing the weights of FNN rather than just the standard technique of back-propagation of errors.