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

T.F. ter Hoeven The NOMAD Operational Controller. A versatile Controller for AGV Systems
Computer program, Report 2005.TL.6958, Transport Engineering and Logistics.

The increasing number of logistic movements and increasing time pressure for those movements create the need for automation of transport systems. To expand the capacity of existing transport solutions and increase the productivity of future systems, flexibility is a key issue. Automated Guided Vehicles (AGV's) are suggested as one of the solutions for transport systems that have to provide reliable, efficient and cost-effective transport movements. To increase the efficiency and flexibility of AGV systems, research is done on new methods of control. A part of this research focuses on the NOMAD model, which is a model for pedestrian behaviour that simulates pedestrian path choice and interaction, created by Serge Hoogendoorn, a researcher at the Faculty of Civil Engineering and Geosciences of Delft Technical University of Delft.

The NOMAD model can be divided into three levels: strategic, tactical and operational. The strategic level defines a schedule of actions or activities; the tactical level uses this activity schedule and information on pedestrian density, time pressure and stimulating effects of the environment to create a velocity field which acts as an input for the operational level. At the operational level, the velocity field serves as a map to direct the pedestrians to their destinations. While travelling, collisions with other pedestrians or obstacles are avoided by control action at the operational level. The first step to application of the NOMAD model on AGV systems is the adaptation of the operational level to create an independently usable controller for AGV's that is able to avoid collisions between AGV's and obstacles or vehicles. The report describes the adaptation of the NOMAD operational level, and shows how the NOMAD Operational Controller (NOC) was modelled.

For two objectives, tests were defined to tune and test the performance of the controller. The first objective was to find controller settings for which the AGV would follow a prescribed path as accurately as possible, to compare its performance with an existing controller which was designed earlier by A. Parajuli, a former student at the department of Transportation Engineering and Logistics. The second objective was to find a set of parameters for which collisions with obstacles and vehicles would be evaded properly, and for which a prescribed path is would still be followed with a reasonable accuracy. The tests were performed in an emulation environment that emulates the control of the scale-model AGV's which are available at the department's AGV laboratory. Tests proved that when tuned for accuracy, the NOC could achieve the same order of accuracy as the Parajuli controller. For lateral control, the NOC performed slightly better. Due to the fact that the NOC is solely a proportional controller, a steady state error occurred for longitudinal control, which means that the AGV is always a small distance behind on its reference path point. Since the Parajuli longitudinal controller is a proportional integral controller, it does not have this problem. However, the step response of the Parajuli controller has a considerably longer settling time than that of the NOC.

The most interesting aspect of the NOMAD-principle is its ability to take evasive action when an object is encountered. Using a simple obstacle evasion test and testing encounters with other AGV's at several angles, a set of all-round settings was determined, for which most situations were dealt with adequately. Obstacle evasion has proven to be very reliable and predictable, and shows that using the NOMAD-principle for AGV control has certainly been worth the effort. Obstacles of all shapes and sizes can be avoided efficiently by the controller, as long as the path to be followed is relatively simple. The vehicle evasion behaviour can still be improved by applying a better traffic decision model than the relatively simple scheme that has been developed so far. The vehicle evasion tests done did show that using the NOC, traffic can be regulated by simply indicating which vehicle should give way, and whether it should pass in front or behind. The controller is flexible enough to ensure a safe distance between the vehicles.

Reports on Transport Engineering and Logistics (in Dutch)
Modified: 2005.07.07; , TU Delft / 3mE / TT / LT.