Driving simulation platform applied to develop driving assistance ...

www.ietdl.org Published in IET Intelligent Transport Systems Received on 4th March 2009 Revised on 21st December 2009 doi: 10.1049/iet-its.2009.0041

ISSN 1751-956X

Driving simulation platform applied to develop driving assistance systems W. Jianqiang L. Shengbo H. Xiaoyu L. Keqiang State Key Laboratory of Automotive Safety and Energy, Tsinghua University, Beijing 100084, People’s Republic of China E-mail: [email protected]

Abstract: This study presents a driving simulation platform with low cost for the development of driving assistance systems (DAS). The platform uses a combination of two simulation loops: hardware-in-the-loop (HIL) and driver-in-the-loop (DIL). Its hardware consists of a simulation computer, a monitor computer, a vision computer, DAS actuators and a car mock-up. Its main software includes a monitor software running in the monitor computer, a vision rendering software running in the vision computer and Matlab/Simulink-based simulation model running in simulation computer. When designing its monitor software, a graphical user interface driven-by S-function method is adopted to eliminate the delay in the displaying of the simulation data. The vision rendering software uses a parametric adjustment method based on the principle of optical projection, improving the driver’s perception of being immersed in the virtual traffic scene. The application of the developed platform is demonstrated by HIL experiments of vehicle actuators and DIL experiments of adaptive cruise control (ACC) algorithm. These experiments not only demonstrate the potential merit of the platform of speeding up DAS development, but also illustrate that the proposed control algorithms for actuators possess good tracking capability, as well as that the developed ACC algorithm is capable of improving driver comfort and reducing driver workload.

1

Introduction

Driving assistance systems (DAS), such as forward collision warning (FCW), lane departure warning (LDW), adaptive cruise control (ACC) and lane keeping system (LKS), are key techniques to further improve driver comfort and vehicle safety of modern vehicles [1]. Nowadays, these systems have been put into use in a few models of commercial trucks and advanced passenger cars, but a number of obstacles still have to be overcome in their mass customisation. One necessary effort is the analysis and testing of sensors, controllers and actuators, determining hardware reliability, algorithm robustness, faulttolerant processing and failure protection. Equally important is the tests’ close connections with traffic environment, driver characteristics and driver’s subjective feelings. Therefore the development must involve numerous field tests with different drivers in various traffic flows. However, these are not only time-consuming and costly, but also of poor controllability and repeatability. In addition, dangers are prone to occur easily in some urgent braking or steering scenarios. IET Intell. Transp. Syst., 2010, Vol. 4, Iss. 2, pp. 121 – 127 doi: 10.1049/iet-its.2009.0041

Driver-in-the-loop (DIL) simulation and hardware-inthe-loop (HIL) simulation are widely used in developing vehicular on-board electronics [2, 3], including DAS. With respect to the evaluation of DAS performance, DIL simulation usually adopts the existing traffic simulator, for example, [4, 5]. The powerful traffic simulator can simulate special testing scenarios, e.g. high speed traffic flow, lowfrictional load, and these special scenarios can be repeated precisely. It can reveal potential bugs in DAS hardware and software effectively and accordingly leads to an improvement in their developing efficiency. Nevertheless, as pointed out in [5], the high cost for constructing and maintaining a traffic simulator, as well as large utilisation expense, restricts its application in the development of DAS. With respect to the operating analysis and testing of DAS hardware, most existing HIL simulation test-beds adopt dSPACE as their simulators, for example, ACC simulator in Hitachi Ltd. [6], ACC simulator in Nissan Ltd. [7] and FCW test-bed in Hanyang University [8]. This type of system is highly integrated and convenient 121

& The Institution of Engineering and Technology 2010

www.ietdl.org for customers, but its cost is unaffordable for most research institutes. Besides, Visual C++ platform is also commonly used in establishing HIL simulator, with relatively low cost. However, the customer programming of simulation codes increases its developing difficulty markedly. To comprehensively address these issues, the paper presents a driving simulation platform based on Matlab real-time simulation environment, which helps to reduce the developing cost and to improve its developing efficiency. It is organised as follows: Section 2 introduces the system configuration, as well as fundamentals of the platform. In Section 3, its hardware is introduced first. The description of its software is given emphasis on the monitor software and vision rendering software. In Section 4, HIL testing of electronic throttle and braking actuator, DIL experiment of ACC performance evaluation are carried out, respectively, in order to validate the primary function of this platform.

2 System configuration of driving simulation platform In general, a driving process can be divided into three stages: perception, decision and action. According to different ways that DAS affect a driver’s action, they can be categorised into two types: one represented by ACC and LKS, the other by FCW and LDW. They both can perform some perception and decision instead of driver. Their difference lies in that the former can further relate actuators as throttle, brake system and steering wheel, thus partly realising automatic

driving. However, the latter does not directly interfere with driver’s manipulation, but provides proper warning information. Therefore when developing the systems above using driving simulator, a HIL simulation function is needed because tests of actuators are important to ACC and LKS. Similarly, a DIL simulation function is needed because driver’s reaction is important to FCW and LDW. For the purpose of being applicable to developing two types of DAS, a driving simulation platform based on Matlab real-time simulation environment is developed. Its configuration is shown in Fig. 1. The configuration contains car mock-up, simulation computer, monitor computer, vision computer, projector, screen, electronic throttle, braking actuator, steering torque imitating device, DAS controller and its corresponding display device. They, together with the driver in the cabin, construct two simulation loops: HIL simulation loop and DIL simulation loop. The car mock-up is a passenger vehicle, of which the engine is removed, but the electronic systems are kept. The former loop is indicated as pink circle (right) in Fig. 1. Fig. 1 aims at testing the hardware of DAS. It consists of simulation computer, DAS controller, electronic throttle, braking actuator and steering torque imitating device. In this loop, the DAS controller receives the virtual information of vehicular sensors given by the simulation computer, for example, longitudinal/lateral states of vehicle, states of power train and so on, and then calculates desired throttle angle, desired braking pressure and desired steering

Figure 1 System configuration of driving simulation platform 122 & The Institution of Engineering and Technology 2010

IET Intell. Transp. Syst., 2010, Vol. 4, Iss. 2, pp. 121– 127 doi: 10.1049/iet-its.2009.0041

www.ietdl.org wheel angle. After sending the desired values back to the simulation computer, the corresponding actuators, including electronic throttle, braking actuator and steering torque imitating device, are controlled and their actual states are acquired by A/D in simulation computer. They construct a closed-loop simulation process.

3.1.2 Vision computer: The vision computer supports vision rendering software. Through RS232 serial bus, it receives the kinetic information of the host vehicle and its surrounding vehicles. Subsequently, vision rendering software, based on MultiGen Creator and MultiGen Vega [9], generates virtual traffic scene, and then projects it on the screen.

With respect to DIL simulation loop, shown as the blue circle (left) in Fig. 1, it is composed of car mock-up, simulation computer, vision computer, projector, screen and test driver. In this loop, the simulation computer collects driver’s manipulating signals through control area network (CAN) bus connected with the car mock-up. Subsequently, the simulation model calculates the states of the host vehicle and those of its surrounding vehicles, and then sends them to the vision computer through RS232 serial bus. By utilising this information, the vision rendering software in the vision computer generates virtual scene and projects it on the screen, to imitate driver’s view in an actual driving process.

3.1.3 DAS actuators: Electronic throttle, braking actuator and steering torque imitating device are three critical actuators of DAS. The first two adjust throttle angle and brake pressure, and the third helps to acquire good manoeuvrability. In order to test the performance of DAS actuators, an electronic throttle from a 1.6 l gasoline engine is installed in the engine cabin of the car mock-up. To automatically control brake pressure in a brake system, an electrical assistant brake device is equipped as braking actuator [10].

In this platform, the two loops are mutually dependent and run simultaneously. They together realise functions such as testing of DAS hardware, driver characteristics experiments, verification of control algorithm and subjective evaluation of DAS performance. Their simulation data are uploaded through Ethernet to the monitor computer, displayed and recorded in real-time by the monitor software. Compared with the existing driving simulators, this configuration possesses simple structure, strong modularity and good maintainability, overcoming such shortcomings as excessive structural complication, high developing difficulties and cost.

3 Design of system hardware and software 3.1 System hardware of platform The following passages focus on the introduction of such hardware as simulation computer, monitor computer, vision computer and DAS actuators.

3.1.1 Simulation computer and monitor computer: The simulation computer is crucial to HIL simulation loop and DIL simulation loop. The xPC target technique is employed to establish its real-time simulation environment. It, based on Matlab real-time workshop, compiles the Simulink model into the executable code automatically and implements real-time simulation in an arbitrary computer, thus dramatically reducing the cost. In the developed simulator, the principle hardware is an industrial computer with CAN card, A/D card, RS232 port and Ethernet port. An xPC real-time kernel forms its software, supporting real-time running of the simulation model. The monitor computer is based on Matlab environment. It supports monitor software, and is able to display and record the simulation data in real time. IET Intell. Transp. Syst., 2010, Vol. 4, Iss. 2, pp. 121 – 127 doi: 10.1049/iet-its.2009.0041

Nowadays, there are three schemes in imitating steering torque resulting from steering system and road feedback: spring mechanism, hydraulic servo-mechanism [11] and electrical servo-mechanism [12]. Considering that the first two have shortcomings such as too fixed feedback resistance and possible deviation from the real steering characteristics, the platform adopts electrical servo-mechanism to imitate the steering torque. Its structure is shown in Fig. 2. Experiments indicate that it can imitate aligning torque and resistance torque of steering wheel acted by road, thus realising manoeuvrability like actual driving.

3.2 System software of platform The key software of this platform includes monitor software and vision rendering software.

3.2.1 Monitor software: As required by DAS, monitor software should have four functions, specifically, parameter adjustment, simulation control, real-time display and data record. Among them, real-time display is the most difficult, because xPC-based simulation computer is not suitable for displaying real-time simulation data. Therefore the monitor computer should perform it instead. To maintain real-time characteristic of displaying data, the sampling frequency of monitor software must not be less than 10 Hz considering the suspenopsia of human eyes. However, experiments show that data displaying will have apparent stagnations if driven by Matlab clock, since graphic user interface (GUI) is an

Figure 2 Structure of electrical servo mechanism 123


www.ietdl.org interpreted language and runs slow. To address the issue, the paper develops the monitor software by a proposed GUI driven by S-function (GUIDSF) method [13]. Fig. 3 shows its basic structure. In GUI, active-X controls are used to achieve virtual metres and real-time display. Simulation data were output from xPC target module in Simulink to S-function, and then S-function drives GUI to display them. Its fundamental is to acquire simulation data by an xPC target block via Ethernet and then drive the active-X controls in GUI by an S-function block. The realtime simulations proved that the sampling frequency of the developed monitor system reaches 100 Hz, the updating frequency of virtual instruments exceeds 20 Hz and the data transmitting and displaying delay is less than 21 ms [14].

3.2.2 Vision rendering software: A critical issue of virtual traffic scene is its weak ‘immersion sense’. Apart from its approximation to reality, the depth of field, size and position of objects are also major causes. They are closely related with viewpoint position and view angle range of vision rendering software, positions of car mock-up, screen and driver’s eyes. Therefore to strengthen the ‘immersion sense’, viewpoint position and view angle range must be

Figure 3 Monitor software by GUIDSF method

Figure 4 Relationship between virtual and real traffic scenes adjusted according to relative positions of car mock-up, screen and driver’s eyes. Here, an adjustment method based on the principle of optical projection is proposed. In order to describe the adjustment method, the presence of a real traffic scene behind the screen is assumed, as shown in Fig. 4. In the figure, A represents the eyes of the driver, B the screen, C the projector, D the farthest point of horizon in the real traffic scene, E the corresponding projected point in the virtual traffic scene, F a vehicle in the real traffic scene and G the corresponding projected vehicle in virtual scene on the screen. To ensure that the driver’s observation from the screen is equivalent to his observation from a real traffic scene, it should be satisfied according to the principle of optical projection that each object in the virtual scene must be the projection of the corresponding object in the real scene when taking driver’s eyes as observation point. For example, line h1 in the virtual scene is the projection of h2 , and virtual vehicle G is that of real vehicle F. For the convenience of calibrating the vision rendering software, traffic elements, for example lane and vehicle, are

Table 1 Adjusting steps Step

Adjustment item

Description

1

screen position

first adjust the screen in its plane until its centre is exactly in front of the driver’s eyes

2

viewpoint position

measure the distance between driver’s eyes and the centre of vehicle plant in platform and then set the viewpoint position of vision generating software to measured value

3

width – height ratio of view angle

set width – height ratio of view angle of vision generating software according to that of the projector device, and ensure that driver’s line-of-sight perpendicular to the screen theoretically

4

projector position

adjust the projector’s pitching angle in order that the farthest point of horizon in the screen scene is at equal height with driver’s eyes. Then implement keystone correction on projector to avoid distortion of virtual scene

5

viewing angle range

calculate the distances from vehicle in real traffic scene to the screen, denoted as d2 and the distance from screen to driver’s eyes, denoted as d1 . Then calculate projection height h1 using the height of vehicle h2 , expressed as h1 =

d1 h d2 2

by adjusting range of viewing angle range, make the vehicle height in the screen scene equal to h1

124 & The Institution of Engineering and Technology 2010

IET Intell. Transp. Syst., 2010, Vol. 4, Iss. 2, pp. 121– 127 doi: 10.1049/iet-its.2009.0041

www.ietdl.org taken as reference objects to describe the relationship between real and virtual traffic scenes as follows: † the farthest point E in the virtual scene is at equal height with driver’s eyes A; † relationship between virtual and real traffic scenes; † the position of vehicle in virtual scene approximates as much as possible that of the projection of its corresponding vehicle in real scene on screen; † the size of vehicle in virtual scene approximates as much as possible that of the projection of its corresponding vehicle in real scene on screen. To meet the conditions above, vision generating software should be adjusted through five steps, as shown in Table 1.

4

Application and verification

To verify the effectiveness of this platform in developing DAS, HIL experiments of actuators and DIL experiments of ACC system are carried out. The success of the platform can be proved if it can verify the control performance of actuators and ACC.

4.1 HIL experiments of actuators To validate function of actuators’ controllers, HIL experiments of electronic throttle and brake actuator are implemented as follows: DAS controller outputs the desired throttle angle ad and brake pressure Pd to the simulation computer; simulation model then executes the corresponding control algorithm, maintaining a preferable tracking of desired values. In the testing of electronic throttle, the desired throttle angle ad is sinusoidal, with amplitude 18% and period 25 s. Fig. 5 shows its control block. The DC motor of electronic throttle is driven by pulse width modulation (PWM) signals, and proportional integral – differential feedback regulates the electronic throttle to track ad . The control algorithm is executed in simulation computer. Fig. 6 shows the experimental results of the throttle angle. It is found that the tracking error is less than 1%. When synthesising a control algorithm for electronic throttle, the HIL simulation results are compared with the demands of DAS. If it satisfies the requirements, the subsequent procedures can be carried out, for example, code programming of control algorithm. Or, a failure-trial process repeats until a satisfying tracking performance of throttle angle is achieved.

Figure 5 Control block of electronic throttle IET Intell. Transp. Syst., 2010, Vol. 4, Iss. 2, pp. 121 – 127 doi: 10.1049/iet-its.2009.0041

Figure 6 Experimental results of throttle angle a Throttle angle b Error of throttle angle

In the testing of brake actuator, the desired brake pressure Pd is sinusoidal, with amplitude 2 MPa and period 35 s. Fig. 7 shows its control block. It controls the high-speed on – off valves of braking actuator by PWM signal, and makes actual brake pressure Pb track desired one through PI controller. Fig. 8 shows the tracking results of brake pressure. It is obvious that the control algorithm can ensure good tracking capability with tracking error smaller than 0.3 MPa. Like the testing of electronic throttle, the performance of brake actuator and its control algorithm could be improved by repeated tests. Known from HIL experiments of electronic throttle and brake actuator, the developed platform possesses the capability of analysing and testing the hardware. There is no need to develop another test-bed when developing DAS applications, thus both reducing development cost and improving development efficiency. On the one hand, the simulator is capable of performing the testing of actuators, thus revealing any malfunctioning at an early stage of

Figure 7 Control block of braking actuator 125


www.ietdl.org effectively because the number of driver’s manipulation decreases. Compared with field test, DIL simulation has great advantages in repeatability and efficiency. On the one hand, the same scenarios of the preceding vehicle can hardly be implemented for both driver manipulation and ACC, thus repeatability cannot be achieved; On the other hand, tests in simulator will dramatically reduce the number of costly and time-consuming field tests. ACC is obviously useful for improving driver comfort and reducing driver workload. The DIL experiment of ACC demonstrates that the developed platform is capable of evaluating DAS performance, thus reducing the number of time-consuming and costly field test, improving development efficiency and reducing development cost, especially during early stages of DAS development.

Figure 8 Tracking results of brake pressure a Brake pressure b Error of brake pressure

development; on the other hand, comprehensive testing of DAS applications with actuators in the loop is particularly helpful to reflect the characteristic of interaction between the driver and the system. For instance, when designing an ACC system, much attention is given to the dual-mode actuators. The term dual mode is used to indicate that the driver can operate the acceleration and braking pedals without intervening with the control algorithm of the ACC. In a traditional HIL simulator, it is impossible to test this function because of the absence of drivers.

To summarise, in the developed simulator, great emphasis has been given for improving the development of DAS, for example, FCW, ACC, LDW and LKS. When developing such systems, the ultimate assessment of system performance should be done by drivers, and the evaluation result is then strongly dependent on the interaction of the driver with the system. This requires a DIL simulation function. In addition, the hardware of the DAS, namely the ECU and the actuators, affects vehicle safety and a primary test of these components in the lab is very much needed. This requires a HIL simulation function. The effectiveness of the two functions in the developed simulator has been demonstrated in the above mentioned tests. However, the two types of tests should not be carried

4.2 Verification of ACC performance ACC has received great attention because of its potential to reduce driver workload and enhance road capacity [15, 16]. In order to evaluate the effect of ACC on driver comfort and workload, a speed profile coming from field tests is adopted as DIL simulation scenarios. Then three experiments (denoted as A, B and C) are carried out. In experiment A, only the driver operates the car, whereas in the other two ACC assists the driver to follow the preceding vehicle. In this research, driver comfort is quantified by the average and variance values of the longitudinal acceleration. Three professional drivers are used, denoted as driver no. 1, 2 and 3. Driver comfort is quantified by average of the absolute of longitudinal acceleration. Driver workload is quantified by the number of brake pedal activations. They are presented in Figs. 9 and 10. Fig. 9 indicates that ACC significantly improves driver comfort, since the longitudinal acceleration level decreases markedly, compared with manual operation. Fig. 10 indicates that ACC can reduce driver’s workload 126 & The Institution of Engineering and Technology 2010

Figure 9 Average of longitudinal acceleration

Figure 10 Number of braking pedal’s activation IET Intell. Transp. Syst., 2010, Vol. 4, Iss. 2, pp. 121– 127 doi: 10.1049/iet-its.2009.0041

www.ietdl.org out in isolation of each other. For this reason, when performing the DIL test of the ACC system, the HIL simulation also runs at the same time because the developed simulator consists of two correlated simulation loops. No malfunction is found in terms of actuator software and hardware. From this perspective, the developed simulator is more adequate for the development of DAS applications than a traditional HIL simulator or a driver simulator, which only have the functionality of either a HIL or a DIL simulator.

5

Conclusions

This paper presents a driving simulation platform for the development of DAS. Conclusions are drawn as follows: 1. GUIDSF method-based monitor software eliminates the displaying stagnation of simulation data. Compared with some effects on virtual scene’s reality, the proposed adjustment method for vision rendering software further strengthens driver’s immersion feeling on virtual traffic scene, improving its reliability of disclosing driver characteristics. 2. The platform processes two simulation loops: HIL simulation loop and DIL simulation loop. HIL experiment of actuators and DIL experiment of ACC proved its capability of speeding up DAS development process and cutting down their development cost. They also illustrated that the proposed actuator algorithm possesses desired tracking capability and ACC is capable of improving driver comfort and reducing driver workload.

6

Acknowledgments

We would like to thank all the reviewers for their very constructive and helpful comments. The project is supported by NSFC (No.: 50705047) and Beijing S&T Nova Program (No.: 2007B-060). It is also supported by the Science Fund of the State Key Laboratory of Automotive Safety and Energy, Tsinghua University. The authors would like to thank Dr. Meng Lu for her continuous help and support.

7

References

[1] AUCKLAND R., MANNING W., CARSTEN O., ET AL .: ‘Advanced driver assistance systems: objective and subjective performance evaluation’, Veh. Syst. Dyn., 2008, 46, (1), pp. 883– 897

[4] THOMPSON K. , TONNIS M. , LANGE C.: ‘Using glance behavior to evaluate ACC driver controls in a driving simulator’. Proc. Human Factors and Ergonomics Society, San Francisco, USA, 2006, pp. 2448 – 2452 [5] GUVENC A., KURAL E.: ‘Adaptive cruise control simulator: a low-cost, multiple-driver-in-the-loop simulator’, IEEE Control Syst. Mag., 2006, 26, (3), pp. 42– 55 [6] ICHINOSE M., YOKOYAMA A., NISHIGAITO T., ET AL .: ‘Development of hardware-in-the-loop simulator for adaptive cruise control system’, Trans. Soc. Autom. Engr. Jpn, 2003, 34, (3), pp. 111– 116 (in Japanese) [7] OIDA T., KUBOTA T., TAKEUCHI T., ET AL .: ‘Development of a vehicle dynamics simulation system for adaptive cruise control’. Proc. Sixth Int. Symp. on Advanced Vehicle Control, Hiroshima, Japan, 2002 [8] YI K., WOO K., KIM S., ET AL .: ‘An experimental investigation of a CW/CA system for automobile using hardware-in-theloop simulation’. Proc. 1999 ACC, San Diego, California, USA, 1999, pp. 724 – 748 [9] CHEN T., WEI L.: ‘Driver-vehicle-road virtual simulation with interaction mode’, J. Chang’an Univ. (Nat. Sci. Edn.), 2007, 27, (1), pp. 67– 71 (in Chinese) [10] GAO F., WANG J., HOU D., ET AL .: ‘Electronically assisted braking system for vehicles based on a high-speed on/off valve’, J. Tsinghua Univ. (Sci. Technol.), 2004, 44, (11), pp. 1532 – 1535 (in Chinese) [11] GAO F., LIU Y., JI X. , ET AL .: ‘Primary calculation of electronically controlled hydraulic power steering’, J. Beijing Univ. Aeronaut. Astronaut., 2007, 33, (5), pp. 605– 607 (in Chinese) [12] WANG T., LU W.: ‘Controller design of automobile electric power steering system’, Comput. Meas. Control, 2008, 16, (3), pp. 330– 335 [13] LI S., WANG J., LI K.: ‘xPC technique based hardware-inthe-loop simulator for driver assistance systems’, Chin. Mech. Engng., 2007, 18, (16), pp. 2012– 2015 (in Chinese) [14] LI S., WANG J., LI K.: ‘Development of real-time monitor system for hardware-in-the-loop simulator’, J. Syst. Simul., 2007, 19, (16), pp. 3684 – 3687 (in Chinese)

[2] YAMAZAKI M., SURESHBABU S., LOFTUS M., ET AL .: ‘Analysis of automatically generated vehicle system control software in a HIL environment’. Proc. of 2002 ACC, Anchorage, USA, 2002, pp. 3135 – 3140

[15] MARSDEN G., MCDONALD M., BRACKSTONE M.: ‘Towards an understanding of adaptive cruise control’, Transp. Res. C: Emerg. Technol., 2001, 9, (1), pp. 33– 51

[3] HU J., ZHAO Y., QIN D., ET AL .: ‘Hardware-in-loop simulation of HEV system based on CAN’, Chin. Mech. Engng., 2008, 19, (3), pp. 300– 304 (in Chinese)

[16] YI K., KWON Y.: ‘Vehicle-to-vehicle distance and speed control using an electronic-vacuum booster’, JSAE Rev., 2001, 22, pp. 403– 412

IET Intell. Transp. Syst., 2010, Vol. 4, Iss. 2, pp. 121 – 127 doi: 10.1049/iet-its.2009.0041

127
