Delft University of Technology
Faculty Mechanical, Maritime and Materials Engineering
Transport Technology

S.A. Röben Position calculation for AGVs using incomplete information.
Computer program, Report 2007.TEL.7124, Transport Engineering and Logistics.

In the AGV laboratory of Transport Engineering of the Faculty Mechanical Engineering at the Technical University of Delft, scale models of AGVs are tested. All Automated Guide Vehicles are equipped with an on-board controller, which controls the drive engines and steering servos. The test field is equipped with and overhead camera system to determine the positions of the AGVs in the field. A computer, using image recognition techniques, decodes the camera images and periodically sends the AGVs their locations via a wireless network.

Problem statement
The allocation of the AGV, from the camera snapshot through the processing of the images to the transmission of data over the network, takes a certain amount of time. This time-delay causes a deviation between the measured location and the actual location of the AGV, which leads to a less accurate controllability of the AGV.

This paper describes the research for a concept to improve the accuracy of position-data received by the AGV.

Principle for compensation
The accuracy of the received position-information will be improved by using the dead-reckoning principle for compensating the position: The updated position-information, the AGV periodically receives, can be seen as the most recent, known position of the AGV. This position is used as a starting point from which an extrapolation of the AGVs movement is calculated, on the basis of actual speed and heading.

The extrapolation of the AGVs trajectory is done by simulation. A vehicle model of the AGV is used for this purpose. This model includes the geometric properties of the AGV plus the electrical en mechanical properties of the actuators. All parameters have to be accurately calibrated for the corrector model behavior. The vehicle model will be controlled using the same signals that the onboard controller of the AGV uses to drive the actuators. Therefore these controller signals are stored for future reference.

The results of the extrapolation of the AGVs trajectory are predicted to give a more accurate representation of the actual location of the AGV at a given moment. The algorithm that calculates the extrapolation will be added to the software of the on-board controller, and is designated in this report as the Corrector.

To further investigate the implementation of the described principle, a TOMAS/Delphi simulation model is written. In this simulation environment one AGV, the positioning system and the Corrector are simulated. The AGV is programmed to run a specified trajectory. The positioning system 'reads' the position of this AGV at a predefined interval. It holds this position for some time, to simulate the time-delay, before it offers the position-data to the Corrector. The corrector then uses this information to extrapolate the trajectory. The deviation between the extrapolated trajectory and the real actual position of the AGV is the key performance parameter for the Corrector.

Several experiments have been done to test the Corrector. The first tests are for the purpose of verification of the simulation model. The following experiments are done to test the influence of variations in some parameters on the behavior of the model.

An important factor is the level of accuracy in which the time-delay of the positioning system can be estimated. Currently, this accuracy is not known for the laboratory setup. This factor is therefore ignored by setting time-delays to a known constant. There are no experiments done to analyze the sensitivity of the model for this factor. The last factor is the operating frequency of the Corrector and positioning system.

Conclusion and recommendations
From the experiments it can be concluded that with the right settings, the corrector is able to improve position information. Because of the fact that all experiments were conducted within a simulation model, it is not possible to make a statement on the improvement that this corrector system will bring when implemented in AGV laboratory environment. It is possible to draw some conclusions on the sensitivity of the corrector for deviations in certain parameters: The correctness of the estimation of the initial state of the AGV at the beginning of an extrapolation trajectory has an influence on the overall performance of the Corrector. Using the steering signals that were given before the positioning snapshot could decrease this effect.

The vehicle model, especially the model of the drive motors, is in the current form very sensitive for errors. This model will have to be calibrated very carefully to achieve good Corrector results.

To enhance the correction method it is recommended to further investigate the possibility of using more position data than just the most recent one. The combined weighted average of multiple extrapolations could then be used to give an improved result.

Reports on Transport Engineering and Logistics (in Dutch)
Modified: 2007.01.29; , TU Delft / 3mE / TT / LT.