Michal Tordon

Covariance profiling for an adaptive Kalman filter to suppress sensor quantization effects

This paper presents a generic approach to model the noise covariance associated with discrete sensors such as incremental encoders and low resolution analog to digital converters. The covariance is then used in an adaptive Kalman filter that selectively and appropriately carries out measurement updates. The temporal as well as system state measurements are used to predict the quantization error of the measurement signal. The effectiveness of the method is demonstrated by applying the technique to incremental encoders of varying resolutions. Simulation of an example system with varying encoder resolutions is presented to show the performance of the new filter. Results show that the new adaptive filter produces more accurate results while requiring a lower resolution encoder than a similarly designed conventional Kalman filter, especially at low velocities.

An active axis control system for a conventional CNC machine

A non-invasive approach to implement an active axis controller for a conventional CNC machine is presented. To develop the system a two-stage approach has been taken. In stage one, a generic external axis controller is designed to bring the machine under the real-time control of an external computer. The external controller is interfaced to the CNC machine via the Manual Pulse Generator (MPC) inputs. The external controller uses a special interface to read the machine position. Stage two is the design of the active axis controller. To achieve this, a sensor feedback loop is added to the external controller to enable visual servoing. The developed active axis controller is used to actively track a profile using Visual feedback. The results presented compares the speed of response of the external axis controller with that of the speed of response of the original CNC controller in linear interpolation motion. The accuracy in the performance of the active axis controller is demonstrated by presenting the results of active visual tracking.

Line Segment Based Scan Matching for Concurrent Mapping and Localization of a Mobile Robot

This paper describes a new approach to align the laser scans in order to build a consistent map of the environment of a mobile robot. There are no physical landmarks and the environment is completely unknown. Our method finds the nearest matching lines in the scans and resolves the data correspondence problem. The global consistency and final map is achieved by the formulation of geometrical constraints and solving them using conjugate gradient method. The method can handle the multiple loop closing problem. This is easy to implement and fully integrated with a complete navigational architecture. The proposed algorithm is implemented using C++ programming libraries and tested in a real-time robotic simulation environment.

In-vehicle data logging system for fatigue analysis of drive shaft

The work presents a cost-effective custom-made data logging system for in-vehicle use. Based on the required bandwidth and accuracy of the torque signal a real-time data acquisition system for long-term data logging of torque signal has been designed and implemented. The system is based on a computer running Windows CE 3.0 operating system. Filtered input signal is sample by over-sampling data acquisition system. Data reduction is achieved by decimation and data compression. Three data compression methods are compared: two methods based on zero-order and first-order predictors and a piecewise linear approximation method followed by run-length and Huffman coding. Several error bands have been investigated in the data compression methods. Test result show that a system with 1 GB flash card can store over 10000 hrs of drive shaft load history data allowing signal reconstruction with satisfactory accuracy.

Curvature based hand shape recognition for a virtual wheelchair control interface

This paper presents a curvature based hand shape recognition method using images of a hand. The curvatures extracted from hand shape contours are converted to a non-dimensional quantity. Combining the contour geometry with the non-dimensional quantities, signatures are generated for template hand shapes. These template signatures are used to recognize hand shapes embedded in newly acquired image contours. The method provides a means of identifying a hand shape in a position and orientation independent manner. Nevertheless, the position and orientation of the hand gesture can still be obtained. The paper then presents a way to use hand shapes in an efficient virtual interface for a wheelchair. The presented system uses only four ergonomically chosen hand shapes to effect command inputs. The wheelchair operates in two different modes: the manual mode and the autonomous mode. The hand shapes used are arranged in a hierarchy taking into account the safety, ergonomics and reliability of operation. Results are presented to show the behavior of the wheelchair in both modes.

A Procedure to make the Probabilistic Odometry Motion Model of an Autonomous Wheelchair

We present a new and completely probabilistic approach to extract the unknown motion model of an autonomous wheelchair. This method is, however, general and can be applied to any mobile robot. Particle filter based methods are very popular for localization and navigation of mobile robots. These methods use motion model for the prediction of robot states when a control action is taken. Well theoretical foundation has been established but only few applications exist. Mostly this model is assumed arbitrarily and fails during the implementation phase. We make the model using probabilistic procedure after getting the real data. The inferences are efficiently made by two table lookup commands during implementation of the particle filter. This can also be used in a hierarchical fashion. The model is validated and experimental results are shown

A hand gesture based virtual interface for wheelchair control

A reliable system that uses hand gestures to control a wheelchair is presented. The control actions are generated by gestures of a bare or gloved hand. The gestures are recognized and identified in a scale, position, orientation and skin color independent manner. Yet the position and orientation of the hand gesture for motion control are used for speed and steering angle control. The interface works equally well with either of the hands and can be interchanged anytime without any changes to the system. The hand gestures are organized in a hierarchy taking into account the ergonomics, reliability of identification and the mode of operation. The modes of operation are; the manual mode and the map mode. The results presented show the behavior of the wheelchair in response to manual and map mode hand gesture commands. The system has been put to test by persons of different hand shapes and proved to be extremely reliable.

Modelling HEUI injector In MATLAB Simulink

The present work describes the development and partial validation of a mathematical model of a hydraulically actuated electronically controlled unit injector (HEUI). The HEUI analyses include submodels of the solenoid, hydraulic differential valve (HDV), intensifier and injector subsystems. It has been implemented using the MATLAB/SIMULINK graphical software environment. The modeled HEUI is a compact, flexible diesel injector developed at the University of New South Wales in conjunction with local industry. The work undertaken is part of a wider study aimed at optimization of the design of the HEUI for dual-fuel systems.

Geometric parameters of hand gesture images for wheelchair control

Geometric properties are used to reliably identify gestures of a bare or gloved hand. A set of optimum hand gestured were experimentally identified. The hand gestures are used to issue commands to a wheelchair. The effect of skin color is eliminated by using a back-lit surface. The system is scale, position and orientation invariant. Yet position and orientation of two hand gestures are used to issue motion control commands in two modes; manual and auto. In manual mode the position and orientation of a selected hand gesture are used for speed and steering control. The interface works well with either of the hands and can be interchanged anytime without any changes to the system. The hand gestures are organized in a hierarchy taking into account the ergonomics, reliability and the mode of operation. The manual and auto mode operations are demonstrated by presenting the results showing each mode of operation.

Visual scanning for off-line automatic tool path generation for shear spinning

Use of visual scanning to automatically generate an accurate tool path is presented. Emphasis is on combining low resolution vision with visual control of a precision machine tool to attain the accuracy required for shear spinning. A simplified edge detection method is used to obtain the required sub-pixel accuracy of the mandrel profile. Two different tool paths are generated using two different methods. First, a part program is generated assuming all data points extracted during scanning are joined by straight lines. Second, assuming that all segments of the mandrel profile consists of circular sections and straight line segments, a curvature detection algorithm is used to identify segments. Then the segments are put together in a part program. Dimensional accuracy of components spun using these two methods are compared with the actual profile. While both methods gave acceptable results, the segment extracted tool path produced a component with superior dimensional accuracy.