Assessment of Advanced Vehicle Control Systems with the vehicle oriented Traffic Simulation Tool PELOPS

to be published: 11. August 1998
1998 SAE FTT Future Transportation Technologie Conference

Copyright © 1998 Society of Automotive Engineers, Inc.

---

Dirk Neunzig, Michael Weilkes
Institut für Kraftfahrwesen Aachen

Almut Hochstädter, Jens Ludmann
Forschungsgesellschaft Kraftfahrwesen Aachen

---

ABSTRACT
INTRODUCTION
THE SIMULATION TOOL PELOPS
VEHICLE MODEL
DRIVER MODEL
ANALYSIS OF ADVANCED VEHICLE CONTROL SYSTEMS
PLATOON SITUATIONS
CONTROLLER PERFORMANCE IN TRAFFIC
CONCLUSION
REFERENCES
DEFINITIONS, ACRONYMS, ABBREVIATIONS

---

ABSTRACT

At the Institut fuer Kraftfahrwesen Aachen, ika (Institute of Automotive Engineering, Technical University Aachen, Germany) the submicroscopic traffic simulation program PELOPS (Program for the dEvelopment of Longitudinal micrOscopic traffic Processes in a Systemrelevant environment) was developed in co-operation with the BMW AG (1). PELOPS is oriented towards the fundamental elements of traffic, namely route and environment, driver and vehicle. The focal point of latest research with PELOPS was the investigation of optimizing measures through the application of new vehicle- and traffic-technologies. Effects on driving-comfort, driving safety, traffic efficiency as well as fuel consumption and emissions are simulated and assessed in urban traffic, various traffic situations on German highways and synthetic testing-procedures. The paper gives information about the simulation techniques and the results of the current analysis.

INTRODUCTION

Current research leads to a reorientation from an isolated view of the vehicle to the vision of the 'intelligent vehicle' on an 'intelligent road'. Thus a variety of systems concerning the influence on traffic were developed in the frame of the European research programs DRIVE, PROMETHEUS and the current German program MoTiV. These systems refer to collective (concerning the complete traffic) as well as to individual (concerning the single vehicle) methods to influence traffic. For an early assessment of these advanced vehicle and traffic control systems the simulation of traffic flow is used. It allows, by means of a virtual picture of the traffic, to give inexpensive, secure and reproducible statements concerning the influence on the course of traffic.

THE SIMULATION TOOL PELOPS

PELOPS was developed at the Institut fuer Kraftfahrwesen Aachen in co-operation with the BMW AG. The idea of PELOPS is a combination of high detailed sub-microscopic vehicle- and microscopic traffic technical models, that permit investigations concerning the longitudinal dynamic of vehicles as well as an analysis of the course of traffic. The advantage of this combination is the opportunity to take all interactions into consideration that occur between driver, vehicle and traffic. An important basis for the realization of this idea is the fact, that the computer capacity was significantly optimized during the last years. Without this capacity the required degree of detail with a simultaneous consideration of all influencing factors would be unthinkable. PELOPS is orientated towards the fundamental elements of traffic, namely route and environment, driver and vehicle. The route model is based on the presentation of the altitude profile with gradients, further on the presentation of the curves with straight stretches of road, arcs of a circle and transitions from a straight route to a curve, as well as of the number of lanes with the respective lane widths. In addition to the geometrical course of the road, the sign postings and the environmental conditions define the state of the route. The route-model covers the entire range from motorways to urban roads, including for example intersections and traffic-lights. The marginal conditions of the traffic situation result from the instructions concerning the number of vehicles that drive on a certain part of the route with a defined length (traffic density) as well as from the starting speeds and the distances between the vehicles (traffic flow). To produce certain courses of traffic or to instruct vehicles with calculated load profiles, single driver-vehicle-units may also be moved according to specific driving speed profiles. In regard to investigations on single vehicles this can be realized by means of standardized driving cycles, like [EUDC], [FTP-75] etc. or in traffic investigations by means of e.g. breaking in panic or constant driving. Figure 1 shows the four main modules of PELOPS and the principle interaction between the models.



Figure 1: PELOPS main structure

VEHICLE MODEL

The vehicle model is based on the 'cause- and effect-method' (2), which means that the order of calculation is based on the operating point of the engine (speed of rotation rate and load) and continuous over the clutch, transmission and differential to the rotor gears where the tractive- and resistance powers are balanced. (Fig. 2) The operating point can be changed by altering the load (cause), that is adjusted by the driver and that leads to a change in power and therefore to a change of the speed of rotation rate (effect). The behavior of the engine is described by means of characteristic maps including e.g. data about the engine's torque, fuel consumption or emissions.



Figure 2: PELOPS vehicle model

The implemented transmission designs are a conventional manual- and an automatic transmission model with a hydrodynamic converter. At the transmission output a retarder model for the simulation of trucks is available. The presentation of the vehicle in this manner according to the cause- and effect principle makes the analysis of autonomous vehicle systems (adaptive cruise control, ABS, etc.) possible.

DRIVER MODEL

The driver model presents the connecting link between the mere vehicle- and the traffic simulation. It is divided into a 'decision-' and a 'handling level'. On the decision level a driving intention is determined based on the driving condition and the traffic environment and consists of acceleration, choice of lane and gear and a strategy-level for reacting to intersection, traffic-lights etc. To ascertain the driving intention, PELOPS works with a psycho-physical distance model, which divides different ranges of driver's behavior by means of reception thresholds. On the handling level the respective intention of the driver is calculated in the corresponding positions of control elements. In this respect the accelerator- and brake pedal are united into a drive pedal. The drive pedal is controlled by a PI-control algorithm. The change of gear takes place with time and torque control; in this case every driver has individual times for the gear change. During lane change the lateral dynamic is neglected. The vehicle is moved alongside a sinusoidal curve from one lane to the other. The shape of the curve depends on the kind of lane change, which is again individually calculated for the driver and the vehicle.

Following Behavior

The basis of the driver model is a psycho-physical follow the leader model. It has been introduced 1974 by Wiedemann (3). For the application in PELOPS this model had to be developed further and extended substantially (4,5,8). In PELOPS the driver is described by typical parameters, such as reaction time, level of perception, level of attention, the need for safety etc. Furthermore the model used in PELOPS distinguishes between different driving situations depending on the surrounding traffic situation and the environment.



Figure 3: Driving situations and levels of perception

The parameters of the driver model are adapted to four different driving situations in which drivers behave significantly different: uninfluenced driving, approaching, braking in emergency situations and following. Depending on the actual vehicle speed, the distance and the differential speed to the preceding vehicle the PELOPS driver model calculates an individual desire for acceleration or deceleration (Fig. 3). As soon as the individually desired following distance is reached, the driver model switches to following mode. A limited control of the acceleration pedal, e.g. due to lack of concentration or driver-errors in gap estimation, leads to distance variations between the minimum and the maximum 'following distance' (see Fig. 3, Dxmin/Dxmax). The described PELOPS model of the driver's behavior is used during 'standard driving situations'. Depending on different traffic situations (e.g. approaching intersections or driving in a traffic jam), the driver's behavior has to be modified to reach the demanded model-accuracy. Therefore a so called model of 'tactical driver's behavior' was developed which adjusts the driver model to different traffic situations.

Tactical Behavior

To adjust the driver's behavior, the actual traffic situation has to be analyzed in a similar manner as the human driver does. That means characteristics like range of visibility or reaction to surrounding vehicles have to be implemented. Figure 4 shows an example of a traffic situation which is analyzed by the 'tactical driver model'.



Figure 4: Detection of traffic situation

The analysis takes e.g. traffic-signs, reduction of the number of lanes or overtaking vehicles into account. Based on this, an adjusted desire for accelerating or lane-changing is calculated. Figure 5 shows the basic principle of the modeled lane-changing decision.



Figure 5. Structure of the lane-changing model

For a various number of lane-changing situations (normal driving, merging, ending of a lane etc.) the parameters of the model are adjusted differently to consider the varying driver's behaviors. Depending on the driver's contentness on their current lane a desire for lane-changing is determined. If the gap in the neighboring-lane is large enough, the driver model will initiate a lane-change. The gap is dependent on the differential speed between the vehicles and the driver's individual need for safety. PELOPS offers highly detailed models for driver, vehicle and environment and draws in this way a virtual picture of real traffic. The accuracy of the calculated results were validated for various situations in urban traffic (6) as well as in highway-traffic (4,7). Combining the driver's model with a vehicle model, based on the principal of 'cause and effect', enables the comprehensive analysis of advanced vehicle control systems like ACC (Adaptive Cruise Control) (see Fig. 6).



Figure 6: Adaptive Cruise Control (ACC) (PROMETHEUS)

ANALYSIS OF ADVANCED VEHICLE CONTROL SYSTEMS

The development of ACC-systems has started more than fifteen years ago and has now reached nearly series-production readiness for highway-applications. During the European PROMETHEUS-program, simulation was a major tool to asses different ACC-systems. Questions about platoon-behavior, platoon-stability and impact on traffic-flow as well as fuel consumption or emissions have been answered by the use of simulation (7). Previous developments concentrated on highway systems because of the reduced number of situations to be handled by the ACC-systems and the great possible increase in driving comfort. Current development focuses on systems which cover the entire range of situations from stop-and-go to complex urban traffic. Figure 7 shows the hierarchical design of a conventional ACC-controller for highway application. The hierarchical design is made by a chain of single controllers which are adapted to the current traffic situation.



Figure 7: Conventional ACC-controller design

This kind of controller reacts only to one obstacle that is in the line of the car and therefore a sensor's tracking algorithm has to predict the relevant preceding obstacle. This method is adequate for the use on motorways, but not sophisticated enough for an application in urban areas. In these traffic-situations the ACC-vehicle can be influenced by a great number of surrounding obstacles. Therefore in the national German MoTiV Program 'ACC im Ballungsraum' a new approach has been chosen. In this program the Institut fuer Kraftfahrwesen Aachen (ika) investigates the characteristics and efficiency of this controller by the use of the simulation-tool PELOPS. The aim of the assessment is to give the controller the most realistic environment possible and to have the ability to make a comprehensive analysis in the early development stages. Therefore, PELOPS offers not only a realistic traffic and vehicle surrounding but a highly-detailed model for the detection geometry, the detection range and even measurement errors of RADAR- or LIDAR-distance-sensors. The assessment is divided into two main parts. In the first step the controller is used in several platoon situations on a one-lane road. In these scenarios the real characteristics of the controller have to be investigated. The criterions for these investigations are platoon-stability, the influence of detection errors on the controller performance, driving comfort and fuel consumption of the single ACC-vehicle. The second part assesses the controller performance in complex traffic situations. In these scenarios the impact of the controller-strategy on traffic-flow, traffic-throughput, safety and fuel consumption as well as emission of the entire traffic is to be shown. In the following a selection of results from the investigations is presented:

PLATOON SITUATIONS

An important criterion for the design of ACC-systems is the platoon stability. In the simulation the platoon stability for the developed control algorithm was determined by means of a dynamic velocity cycle (figure 8), originating from the measurings on a German highway. This velocity cycle was implemented into a vehicle that was being followed by an ACC-sensor equipped vehicle.



Figure 8: Velocity course

Then, on the basis of linearisation and application of MATLAB the Bode-diagram for the controller was prepared by the transmission behavior in the velocity course (figure 9).



Figure 9: Platoon stability in the bode-diagram

Here, it can be detected in the amplitude course that one criterion of the column stability is fulfilled by the damping occurring over the whole frequency range. A further condition for the stability of an ACC-sensor lies in the fact that the velocity course does not show any raise in case of an (unrealistic) impulse by a velocity jump of the leading vehicle. As figure 10 shows this condition is not fulfilled over the whole platoon, because an increase of the maximum speed compared to the preceding vehicle does occur for single vehicles. However, at the same time the overspeeding is reduced again over the further platoon vehicles so that this behavior can be disregarded for the traffic flow.



Figure. 10: Platoon behavior in case of impulse through velocity jump

The already in the amplitude course of the controller traceable relief of the velocity course over the platoon position becomes also clear in the simulation of a platoon of 10 vehicles, in which the first vehicle does again set the above demonstrated velocity course. (figure 11).



Figure 11: Velocity courses of a platoon in highway cycle with an headway of 1.5 sec.

From the relief of the velocity course also results a reduction of the average fuel consumption. Figure 12 shows the percentile change as against the leading vehicle for two different intervals. Additionally, the influence of noisy input data of the controller on the fuel consumption is indicated.



Figure 12: Fuel consumption over the platoon position for different headways

In case of the noisy data the own vehicle data provided to the controller - in particular velocity - as well as the data of the preceding vehicles, given by a sensor model, were altered by white noise and individual errors up to a maximum of 7%. The relevance of the sensor data is well illustrated in the consumption data. The reason for the higher consumption in case of noisy input data lies in a greater scattering of the engine's operating points, as illustrated in figure 13.



Figure 13: Engine operating points for leading and following Vehicles at an Headway of 1.5 s

CONTROLLER PERFORMANCE IN TRAFFIC

In the second part of the analysis the focal point lied on the effect of the ACC-algorithm on the traffic flow. Of the investigated scenarios the effect in inner-city traffic shall be presented here. A highly frequented, traffic light controlled and one-lane access road in the Aachen suburban area served as investigation ground. Here, measuring data was collected in the morning rush hour, in order to register the traffic in terms of macroscopic data (average speed and traffic density, relating to overall traffic) as well as of microscopic data (frequency distribution of distances and velocities, relating to single vehicles). On the basis of this data a simulation with PELOPS was arranged. The parameters were calibrated so that the measured situation could be reproduced. In the following this "Basis-Simulation" serves as reverence for simulations with ACC. Different equipment degrees were analyzed for the effect of the ACC, in order to demonstrate the effects in different time horizons. The equipment degrees were set to 5%, 20% and 40%. Additionally, a simulation with 100% equipment - meaning that only ACC-vehicles were used - was carried out for the assessment of the ACC's potential. An observation of two different controller levels took place differing in the data provided for the control algorithm and the preset intervals. In case of controller level 0 the time interval is set to the values of 1.2, 1.5 and 1.8 s depending on driver characteristics. This controller level does further base on pure sensor data, like those being provided by a radar or lidar distance sensor, whereas controller level 1 can dispose of even further information. This data includes exact lane information, such as the lateral position of the vehicles on the lane, recordable for example by a picture processing system. Additionally, the interval of the ACC is set to 1 s in the simulations of controller level 1. Figure 14 illustrates the results of different simulations in the fundamental diagram. For a better survey only one value middled over the whole course is implemented into the simulation.



Figure 14: Fundamental diagram of different simulations in urban traffic

In case of a raising equipment degree a raising traffic volume as against the basis simulation can be detected for the two controller levels, signifying a raise of throughput. In case of 'total equipment' an increase in speed is additionally registered. Controller level 1 leads, as expected, to a significant improvement in traffic flow compared to level 0 because of the smaller interval and better sensor data. As further simulation variant an investigation of a 'realistic' reproduction of the surrounding situation was done. Now, 'disturbing objects' at the side of the road lying in the registration area of the sensors were added. Because the controller should also react to standing objects, disturbances were purposely implemented. Here, already the worse macroscopic traffic data show that frequent reactions took place to these 'disturbing objects', not lying in the direct driving course (simulation controller level 0 as 'disturbing objects 0, level 1 as 'disturbing objects 1' in figure 14). This renders the requirements clear that have to be fulfilled by the sensors and the driving course prediction, in order to avoid such errors. As last variant a variation of the driver interface was investigated. Whereas in the present simulations the driver had to activate the ACC after every stop, the ACC remained 'on' in the simulations 'starting'. Thereby, the restarting is faster by the specific reaction time of the driver. The results of this simulation can also be found in picture 14. Here, a significant raise of the traffic volume through the automated vehicle start becomes evident.

CONCLUSION

The results demonstrate the performance efficiency of the simulation tool PELOPS. The presented results constitute only a small part of the investigations. Most of all, the influence of the test simulations - particularly the column scenarios and specifically composed single investigations with preset velocity courses of the preceding vehicle - on the further development of the controller algorithm cannot be shown due to the lack of time. By means of the simulation different filter procedures or parameterisation of the controller in different situations can be adjusted in the simplest way. Apart from statements concerning fuel consumption of engine operating points of course, also an adjustment of the remaining drive train to the ACC can be carried out, in particular of the gear changing strategy of an automatic transmissions. Additionally, by means of an analysis of macroscopic situations the effects of a controller on the whole traffic can be presented at a very early stage. For the investigated controller on the one hand a further development and the discovering of weak points were achieved, and on the other hand its efficiency in the whole traffic was proofed. These results were only made possible by the special characteristics of the simulation tool PELOPS. In particular the accurate reproduction of the road- and sensor geometry, the realistic driver modeling as well as the exact calculation of the drive train after the cause-effect principal lead to outstanding realistic results.

REFERENCES

DIEKAMP, R.; LUDMANN, J.; LERNER, G.
PELOPS- Ein Programmsystem zur Untersuchung neuer Längsdynamikkonzepte im Verkehrsfluß, VDI Berichte Nr. 1007, Wuerzburg 1992

DAVID, W.
Modulares Simulationsprogramm zur Auslegung und längsdynamischen Untersuchung von Kfz-Antriebssystemen, Dissertation am Institut für Kraftfahrwesen der RWTH Aachen, 1992

WIEDEMANN, R
Simulation des Straßenverkehrsflusses,Schriftenreihe des Institutes für Verkehrswesen der Universität Karlsruhe, Heft 8, Karlsruhe, 1974

EHMANNS, D.
Modellierung des Fahrerverhaltens bei verschiedenen Verkehrszuständen zur Analyse von Fahrerassistenzsystemen, Diplomarbeit, Institut für Kraftfahrwesen der RWTH Aachen, 1996

NEUNZIG, D.
Erweiterung des Simulationsprogramms PELOPS zur Abbildung von Verkehrsabläufen in Verdichtungsräumen,Diplomarbeit, Institut für Kraftfahrwesen der RWTH Aachen, 1996

LUDMANN, J.; NEUNZIG. D.; WEILKES, M.
Investigation of Intelligent Traffic Systems by Means of Simulation, 4th World Congress on ITS, Berlin: 1997

LUDMANN, J.; WEILKES, M.
Entwicklung, Analyse und Bewertung von PROMETHEUS Konzepten unter besonderer Berücksichtigung von Autonomous Intelligent Cruise Control mittels Simulation, Abschlußbericht zum Eureka Verbundprojekt, Aachen 1995

LUDMANN. J.
Konzeption eines Verkehrssimulationsprogramms
Diplomarbeit, Institut für Kraftfahrwesen der RWTH Aachen, 1990

DEFINITIONS, ACRONYMS, ABBREVIATIONS

ACC Adaptive Cruise Control
DRIVE Dedicated Road Infrastructure and VEhicle Systems
MoTiV Mobilitaet und Transport im intermodalen Verkehr (Mobility and Transport in intermodal Traffic)
PELOPS Program for the dEvelopment of Longitudinal micrOscopic traffic Processes in a Systemrelevant environment
PROMETHEUS PROgraM for a European Traffic with Highest Efficiency and Unprecedented Safety