Fuzzy Longitudinal Controller Design and Experimentation for ...
Fuzzy Longitudinal Controller Design and Experimentation for Adaptive Cruise Control and Stop & Go Chien-Tzu Chen1,Ching-Chih Tsai 2, Shih-Min Hsieh3 1 2,3
Automotive Research & Testing Center, Changhwa, Taiwan, R.O.C.
Department of Electrical Engineering, National Chung-Hsing University, Taichung, Taiwan
Abstract This paper presents a fuzzy longitudinal control system with car-following speed ranging from 0 to 120 km/h. This controller achieves the main functions of both adaptive cruise control (ACC) and Stop & Go control. For constructing the vehicular longitudinal control system, we use vacuum boosters to control the throttle and the braking pedal, thus circumventing the technical difficulties of using engine management system and anti-brake system. A fuzzy controller is synthesized by inputting the difference of the actual relative distance and the safe distance obtained from the preceding vehicle, and the relative speed, and then outputting the pulse-width-modulation (PWM) signal to control the output forces of the vacuum boosters. With the use of the high-level controller from dSPACE, the fuzzy control law is easily and rapidly implemented using Simulink for the experimental car, and the controller’s parameters can be changed and updated by analyzing the collected data from the relative distance from the lidar, the speed of the host vehicle, the opening of the throttle and the position of the braking pedal. Modified by a commercial SUV, the experimental car is equipped with a lidar on the bumper, a speed sensor mounted at the shaft of a wheel, and two sensors for measuring the opening of the throttle and the position of the braking pedal. For the sake of safe testing, experimental results are conducted by virtually simulating the various possible car-following conditions for the ACC and Stop & Go controllers, thereby obtaining virtually relative distance and speed to tune the controller’s parameters and make sure the safety of the controller. Afterwards, the realistic car following experiments are then conducted to confirm that the proposed fuzzy controller is capable of achieving the requirements of comfort and safety and giving a satisfactory performance at high and low speed conditions.. Keywords ： Adaptive cruse control, fuzzy control,
I.
INTRODUCTION
Currently ACC has been seen as a comfort system [1] for use to reduce driver fatigue in heavy traffic, but it is has the possibility to provide a lot of safety related feature in the near future and become an integral part of an intelligent transportation system (ITS). ACC is the longitudinal control for the vehicle dynamics; in the near future it combines the lateral control via active steering system and lane recognition systems or the intelligent transport infrastructure (e.g., high precision GPS), the dynasty of auto-driving will come true. ACC systems have been in market as an optional device for luxury vehicles. They have been developed for highway driving assistance, but could not handle urban traffic situations. One limitation of these commercial systems is that they control the speed of the car only at speeds above 40 km/h and they fail at lower speeds in heavy traffic. The failure may result from two key reasons. One is sensing difficulty in complex urban. The front vehicle situation of cut-in and lane change are more frequency, and the recognition pedestrian must be completely considered. The other is the frequent switching between acceleration and deceleration, and the smooth control for the driving stability and comfortable at low speeds is more difficult than at high speeds. The commercial systems [2] achieve vehicle longitudinal dynamics control via integration engine management system (EMS) and anti-lock braking system (ABS), thus their implementation do not need additional hardware. Considering the requirement of aftermarket and there are no independence for the two key systems at the local automotive manufacturers, we need to design a new type of by-pass actuator to control the vehicle longitudinal dynamics of acceleration and deceleration. In this paper, we desire to develop a longitudinal control method that could satisfy the driving condition between 0 to 120km/h, actually including the functions of ACC and Stop&Go control.
longitudinal control, stop & go. To achieve both functions of ACC and Stop & go, we
have to mount required sensors in the front of the car, measuring the preceding vehicle’s velocity and distance. The sensor could either be of optic or radar type, but the radar sensor is often preferred since it is much less influenced by the weather conditions than the optical sensor. The sensor information is then transmitted to the system’s electrical computer unit that controls the engine and the brake system. For safety reason, the car controllers only allow gentle acceleration and deceleration (under 0.3g) such that the driver should always be able to intervene if the system does not comply with the driver’s intentions. It is important to remember that the system is only a service to help the driver, not a replacement of the driver. Fuzzy logic is a powerful technique; in particular, it permits control without extensive knowledge of the equations of the plants or processes and it represents in a very effective way the human reasoning methods [3], [4]. The conventional PI controller was applied to the Stop&Go system [5]. This kind of controller [5] was divided into two separate parts: the outer control loop generates a desired acceleration that represents the driver behavior and the inner control loop controls the brake pressure and the throttle position that is dependent of the vehicle dynamics. Computer simulations are always helpful and useful in providing an assistance tool to facilitate the development of the ACC and the Stop&Go control, thus reducing development time and removing the risk on testing. The multi-body vehicle simulation tool Fasim_C++ provides an ideal environment to simulate vehicle behavior with ACC [6]. The longitudinal vehicle model has been used to accomplish ACC and Stop&Go control. The longitudinal controller is mainly composed of an engine, a torque converter, a transmission, a drive train, and a brake system. The simulation of the longitudinal controllers with fuzzy logic and neural networks based controller have been shown capable of giving good performance that provides the information of membership design [7]. Traditional cruise control is a universal system today, so acceleration control is not special technique by electronic throttle control. Deceleration Control for braking is typically performed using a “smart” booster, brake-by-wire, or an ABS modulator (hydraulic mechanization). Deceleration control for using a “smart ” booster is done by controlling the air flow valve of the booster, thereby regulating the vacuum inside the booster. Deceleration control using a brake-by-wire implementation has been performed by sending a brake command to the actuator at each wheel, which applies braking to the wheels. Deceleration control using the ABS modulator is implemented by regulation of wheel pressure through solenoid valves within the modulator [8], [9], [10]. ACC systems can provide assistance to the driver in a lot of situations, but never can relieve the drive’s
responsibility for vehicle guidance. The driver always has full responsibility for the driving task. For safety design, it is necessary to limit the systems deceleration and acceleration capability in order to avoid irritation of the driver and the surrounding traffic in case of inappropriate control reaction. In practical use a range od-2.0 m/s2 to +1 m/s2 was found to be a suitable compromise between customer benefit, convenience and safe requirement. It is necessary that the driver is allowed to override the system at any time by throttle or brake it without any conflicts with system functions [11], [12], [13]. Another critical point to a successful ACC system is the high performance sensor that has capability to detect the vehicle ahead of the host-vehicle, and measure the kinematic attributes of the target (e.g. distance, and relative velocity, etc.). For the need of the long sensing distance and reliability, lidar- and radar- based object detection technologies have been widely used. They need fast reaction to the dynamically changing forward traffic behavior (e.g. cut-in, lane changes, etc.) in the presence of various weather conditions (e.g. rain, fog, etc). Their requirement include the update time of less than 100 msec., the measurement range of relation speed from –150 km/h to 150 km/h, and the measurement range of the relative distance more than 100m [14], [15]. The advantage in CMOS imaging sensors has enabled low-cost and high-quality cameras that are making their way into future automobiles. In the future, vision sensors can be deployed in a car to improve robustness of the ACC system [16], [17]. The paper aims at developing a fuzzy longitudinal controller and investigating its feasibility and efficacy through several experiments. The contributions of the paper are fourfold. First, vehicular longitudinal control design is accomplished by using vacuum boosters to control the opening of the throttle and the position of the braking pedal. Worthy of mention is that this design does not involve with the complicated and sophisticated knowledge of the engine management system and the anti-brake system. Second, a single fuzzy controller is designed to carry out three control functions, including conventional cruise control, ACC and Stop&Go control. Third, the parameter tuning of the fuzzy controller is done by employing virtual car-following simulations. This kind of virtual experiment is safe and effective in tuning the controller’s parameters in an off-line manner. Finally, low-speed Stop & Go control can be easily accomplished by the proposed fuzzy control method. In addition, through experimental results, the proposed fuzzy method together with the developed control system has been shown effective and useful in not only achieving high-speed ACC control and low-speed stop & go control, but also fulfilling conventional cruise control. The remaining sections of the paper are outlined as follows. Section 2 details the novel control system architecture and its key components of the experimental platform. In Section 3, a fuzzy control approach is
briefly reviewed and a fuzzy longitudinal controller is presented in detail. Section 4 conducts several experiments to show the efficacy and merit of the proposed controller. Section 5 concludes the paper.
II. SYSTEM ARCHITECTURE AND DESCRIPTION This section aims at presenting new control architecture of the proposed fuzzy longitudinal control system with brake actuation, and describing the key components used for constructing such a system. This new design does not change or modify the original engine management system (EMS) and anti-brake system (ABS), but inserts additional components to achieve vehicular longitudinal control. Unlike the conventional adaptive cruise control systems, the proposed longitudinal control system uses vacuum boosters as main actuators for the throttle and the braking pedal, thus providing better controls to steer the vehicle; the kind of control system with the vacuum boosters is particularly useful in offering more aggressive control actions to brake the vehicle in case of emergent braking of the preceding car. In comparison with conventional actuators made by the direct-current motors, the proposed actuators are able to actuate the throttle and the braking pedal in both forward and backward directions, whereas the conventional actuators driven by the DC motors are only suitable for only one-direction driving. 2.1 Description of the Overall System Structure This subsection is devoted to describing the overall system structure of the proposed fuzzy longitudinal control system. Fig.1 shows the overall system architecture of the proposed control system. Four sensors, including the Lidar, the ABS speed sensor, the throttle position sensor, and brake position sensor, are used to provide all the necessary signals for the control system. The data recorder is responsible for collecting the signals from the four sensors. The data recorder is connected via Ethernet to the first computer, called Computer A, which monitors all the signals in real time. Worthy of mention is that all the signals must be connected to the data recorder For the purpose of real-time feedback control, the two signals obtained from the Lidar and the ABS speed sensor are directly inputted to the high-level controller, the MicroAutoBox 1507,made by dSPACE, thus generating the desired PWM commands to drive the two vacuum boosters.. The loop refresh rate (sampling) is performed once every 1/100 of a second. The second computer, named “Computer B”, is adopted to write and edit the control codes of the proposed control laws for the dSPACE high-level controller, tune the controller’s parameters, and acquire and display the real-time command information via the optical fiber network. The
display unit for driver attains the real-time command information and the current status of the experimental platform from Computer B via RS-232, and shows these messages to the driver. The main actuators are made by the two vacuum boosters, offering the driving and braking forces for the experimental platform. 2.2 Throttle and Braking Actuators This section uses the two vacuum boosters to design the braking pedal actuator and the gas pedal actuator. Fig.2 depicts the cross-sectional view of the vacuum booster. The pressure difference between the two sides of the piston produces a force to pull the gas or braking pedal through the steel rope. The pressure at the upper side of the piston is almost 1 atm because this air room is directly connected to the air. The pressure at the lower side of the piston can be regulated by exciting the three coils in the three electromagnetic valves. The electromagnetic valve B is used to adjust the opening of the valve by its PWM commands, thus reducing the pressure of the chamber inside the piston. Both two electromagnetic valves (A and C) are utilized to increase the pressure of the chamber inside the piston. To simply the control of the vacuum booster, we let the valve A be always closed and control the valve C by using PWM commands. For the sake of simple control, the two valves C and B are controlled by using the same PWM commands. Next, we move to the design of the braking pedal actuator. As shown in Fig.1, the vacuum booster is adopted to achieve the function by directly pulling the braking pedal. Without the pulling force, the braking pedal will be pushed back the original position by its internal spring. When the two forces are equal, the braking pedal will stay at an equilibrium position. Hence, the designed actuator will regulate the position of the barking at any desired location by applying appropriate PWM commands to the vacuum booster. The same design philosophy can be applied to the gas pedal actuator. Similar to the braking pedal actuators, the vacuum booster is directly used to pull the gas pedal, thereby changing the opening of the throttle via the steel rope. As shown in Fig.1, the gas pedal will recover to the original position by its internal spring while releasing the pulling force. Accordingly, the gas pedal actuator will be easily controlled by appropriate PWM commands. 2.3 Experimental Platform The experimental car used in our study was 1997 model-year HONDA SUV. This experimental platform was equipped with an automatic gearbox in order to simplify the gear change. With all the components mentioned before, the experimental car is equipped with the two computers, the data recorder, the dSPACE high-level controller, the display unit for driver, the four sensors and the two actuators, and the power supply.
Fig.3 presents the appearance of the experimental vehicle, where the lidar is mounted on the bumper of the experimental car.
III. FUZZY LONGITUDINAL CONTROLLER DESIGN This section presents a fuzzy longitudinal control method to achieve both high-speed adaptive cruise control and low-speed Stop & Go control. The conventional adaptive cruise control is usually concerned with the control of the host vehicle with speeds up to 40 km/h, in order to keep a safe distance away from the preceding vehicle. In contrast to adaptive cruise control, the stop & go control is often applied to steer the host car with speeds less than 40 km/h, dealing with the car-following problem with a safe distance away from the preceding vehicle. In the section, the proposed fuzzy control method is capable of addressing both two issues in a unified control framework. 3.1 Longitudinal Controller This subsection is aimed to describe the way of how to achieve the longitudinal control and to explain the proposed control logics to accomplish the conventional cruise and the adaptive cruise control. Under normal driver control, the acceleration and deceleration of the vehicle is maintained by use of the throttle alone. The throttle is opened to accelerate the vehicle, but closed to decelerate the vehicle due to engine braking effect. This implies that the longitudinal controller can decelerate the vehicle by making use of either the throttle or the normal braking. For braking, engine braking is used for the small deceleration whereas the normal braking is adopted for the larger decelerations. Although it is important to use the longitudinal controller to carry out precise speed tracking and car-following, the throttle control is still the main driver strategy under the consideration of the driver’s comfort and the fuel consumption. The normal braking control is utilized only in case of emergency, thus avoiding the frequent switching of the throttle control and the normal braking control. Although the cruise control and the so-called adaptive cruise control are done independently, this section proposes a longitudinal controller to accomplish precise speed tracking and car-following. Note that the proposed longitudinal controller works for the speeds ranging from the standstill condition to 120 km/h. Fig.4 describes the architecture of the proposed longitudinal controller. The relative distance via the lidar and the speed of the host car from its ABS speed sensor are considered as two inputs of the longitudinal controller. The relative distance information is taken to judge whether the preceding car appear in front of the preceding car. If there exists a preceding car, then the safe distance between the two cars must be evaluated by
taking into account the speed of the preceding car and the time gap set by the driver. The time gap is referred to the human reaction time from the moment that the driver is aware of danger to the moment that the driver takes actions to avoid the risk of danger. In other words, the time gap means that the minimum safe time to prevent the driver from dangerous conditions. In general, the time gap roughly ranges from one second to two seconds, according to the study of human agronomy. In this experiment we set the maximum of two seconds. In Fig.4, there are two other conditions to calculate the safety distances. The first case is that the safety distance is assumed to be one hundred meters if no preceding cars are found. The second case is that the safety distance B is regarded as the product of the desired time gap and the desired cruise speed specified by the driver. The final safety distance is the minimum of the sate distance A and the safety distance B. The proposed fuzzy longitudinal controller will be activated with the two inputs of the final safety distance and the related speed between the host car and the preceding car. Worthy of mention is that the proposed longitudinal controller will limit the speed of the host car to the desired speed if no preceding cars occur, thus preventing the host car from monotonically increasing its speed to impractical infinity. 3.2 Fuzzy Controller Design This subsection presents the proposed longitudinal control method. Fig.5 describes the block diagram of the fuzzy longitudinal control system. In Fig.5 the outputs of the longitudinal controller are the distance error and the relative speed between the host and preceding cars, which are the inputs of the fuzzy longitudinal controller. The relative speed Vr is calculated from the speed difference of the preceding and the host cars, i.e.,
ΔV = V p − V h
(1)
where Vp and Vh denote the speed of the preceding car and the speed of the host car. The distance error ed is defined by subtracting the relative distance from the safety distance, i.e.,
Δd = D s − D r
(2)
where Ds and Dr denote the safety distance and the relative distance, respectively. It is worthwhile to note that the relative distance is measured by the lidar and the safety distance is obtained from the product of the desired time gap and the speed of the preceding car plus the safety factor, that is ,
Ds = V p * t g + f
(3)
where t g and f denote the time gap and the safety factor, respectively. The safety factor means the offset distance to avoid car collision if the speed of the preceding vehicle becomes zero. In doing so, we set the
sufficiently safe distance of 10 meters in this experiment. After executing the fuzzy control rules specified by the designer, the fuzzy longitudinal controller will generate the actual control signal u. Note that the control signal u is in the interval [-1,1]. If the control signal u is positive, then the control u is transformed into a corresponding PWM command and the gas pedal actuator will be activated according the PWM command. Conversely, if the control signal u is negative, then the braking pedal actuator will provide the braking action, based on the braking PWM signal which is proportional to the absolute magnitude of the control u. Based on the PWM command, the gas or braking pedal actuator will produce a pull force to control the opening of the throttle and the position of the braking pedal, hereby accelerating or decelerating the host car and eventually obtaining the desired speed. In case of no preceding cars, the fuzzy longitudinal control will work in the cruise control mode. In the cruise control, the fuzzy longitudinal controller is used to achieve precise constant speed tracking and speed profile tracking. However, the physical quantities, including the safety distance DS , the relative speed ΔV and the distance error Δd , must be modified as follows.
DS = Vt × t g + f
(4)
ΔV = Vt − Vh
(5)
t
Δd = D s − ∫ ΔV × d τ 0
(6)
where Vt is the target speed specified by the user. Fig.6 depicts the fuzzy sets and the rule matrix of the fuzzy longitudinal controller. The membership functions of the input linguistic variables are triangular, and the output membership function of the control signal u is also triangular, thus simplifying the computations of the fuzzy controller. The two inputs are fuzzified into five linguistic variables, including NL (negative large) , NM (negative medium) , ZE (near zero), PM (positive medium) and PL (positive large), and so is the output u. However, the membership functions of the linguistic variables of the control u are different from those of the inputs. As shown in Fig.6; the linguistic variable ZE of the output significantly overlaps the linguistic variables NM and PM. Such a design eliminates frequent switching of the control signal from positive to negative, and causes the vehicle to move smoothly and comfortably, thereby avoiding abnormal fuel consumption. The fuzzy control rules are constructed based on the driver’s experience, thereby producing simple 25 control rules. These fuzzy driving rules are easily comprehended. For example, if the relative speed is NL, and the distance error is NL, then the control u is NL. This rule implies
that the host car is fast approaching the preceding one and the two cars are very close, then the large braking action should be taken to significantly reduce the speed of the host car in order to avoid car collision. Moreover, the rule matrix is symmetric with respect to the diagonal entries. Last but not least, the fuzzification ranges of the linguistic variables for the two inputs can be adjusted so as to achieve satisfactory accelerations and decelerations of the host car. The modification of the overall fuzzification ranges of the two inputs may also provide better drivers’ comfort. For example, if the fuzzification interval of the relative speed becomes narrow, then the accelerations and decelerations of the host car will turn out large, thereby causing the driver to feel discomfort. Conversely, if the interval of the relative speed becomes large, then the host car will have mild accelerations and decelerations and the driver will feel no discomfort. The situations also hold for the distance error, but the change of the fuzzification interval of the relative speed gives more effective tuning. Thus, the effective fuzzification intervals of both the distance error and the relative speed can be experimentally determined so that the desired accelerations and decelerations are achieved. Through our experiments, the useful fuzzification intervals for the distance error and the relative speed are selected as the interval [-30, 30] and the interval [-10,10], respectively.
IV. EXPERIMENTAL RESULTS AND DISCUSSION This section aims at conducting several experiments to examine the feasibility and effectiveness of the proposed longitudinal control system and fuzzy longitudinal control method. These experiments used the following settings: the desired time gap was two seconds and the safety factor was two meters. Before the performance of the overall system and the proposed control method were tested, it is necessary to investigate the input-output properties of the vacuum boosters and the characteristics of the Lidar. Several virtual simulations were employed to on-line tune the parameters of the fuzzy longitudinal controller in order to obtain satisfactory performance. Afterwards, the simple constant speed command tracking and speed profile tracking experiments of the fuzzy longitudinal controller were performed. Finally, the low-speed stop & go experiment was conducted in a realistic road environment. Fig. 7 and Fig. 8 respectively depict the performance of the proposed fuzzy longitudinal controller for conventional cruise control at 80 km/h and speed command profile tracking with real measurements from the Lidar. These experimental results indicate that the proposed fuzzy longitudinal controller is shown capable of achieving conventional cruise control and speed commands profile tracking. Worthy of mention is that the result in Fig. 7 (c) showed that the braking pedal was not activated for accomplishing the constant speed cruise
control, but in Fig. 8 (c) the throttle complemented the braking actuation for carrying out the constant command profile tracking. This fact indicates that the proposed controller behaved like a human driver.
systems are carried out using two independent controllers. The significant feature of the proposed method hinges on its capability to achieve ACC and stop & go control in a unified control framework.
Afterwards, this type of controller was applied to steer the host car in achieving the goal of adaptive cruise control. The stop & go experiment was setup as follows. The preceding car moved in the speed of 20 km/h and it was far from the host car. The host car moved forward with a speed of 50 km/h. In such a way, the relative distances between two cars were gradually reduced and the proposed fuzzy longitudinal controller mounted on the host car started to work. At the same time, the preceding care took actions of acceleration and deceleration, thus providing the time-varying speed profiles as shown in Fig.9(a). Fig.9(a) shows that the host car tracked the preceding car with small speed errors. Moreover, the result in Fig.9 (b) clearly indicates that the host car followed the preceding car at safe distances, and the result in Fig.9 (c) reveals that the time gaps between the two cars were greater than 3 seconds. As can be observed in Fig. 9 (d) and (e), the throttle worked with the braking pedal in a human driving manner, that is, the throttle complemented the braking actuation so that the proposed fuzzy longitudinal controller performed well in keeping safe inter-vehicle time gap.
Third, the parameters of the fuzzy controller are experimentally tuned by employing the virtual car-following simulations. These simulations take into account various real car-following conditions of the ACC and Stop & Go systems. These virtual simulations have been proven safe and effective in off-line tuning the controller’s parameters. The tuned parameters are used for real ACC and stop & go experiments.
V. CONCLUSIONS This paper has presented techniques to establish a new kind of adaptive cruise control system and to present a fuzzy control method. The main techniques developed in the paper include how to design vacuum boosters for actuating the throttle and the braking pedal, how to set up the lidar, sensors for measuring the angles of the throttle, the positions of the braking pedal and the speed of the host vehicle, how to synthesize the all-speed fuzzy controller, and how to proceed with real car-following experiments. The effectiveness and performance of the proposed system and method have been investigated in detail. The main results of the paper are summarized as follows. First, the vehicular longitudinal control design is accomplished by using feedback signals from Lidar and sensors for measuring the angles of the throttle, the positions of the braking pedal and the speed of the host vehicle, and by employing vacuum boosters to control the opening of the throttle and the position of the braking pedal. It is worthwhile to note that this new design does not change or modify the original engine management system (EMS) and anti-brake system (ABS), but inserts additional components to achieve vehicular longitudinal control. Second, a single fuzzy controller is synthesized to accomplish out conventional cruise control, ACC and Stop & Go control. The type of controller is a breakthrough because the current ACC and stop & go
Finally, low-speed Stop & Go control can be easily accomplished by the proposed fuzzy method. Conversely, conventional approaches are hard to achieve stop & go control in a smooth way of acceleration and deceleration. In addition, through experimental results, the proposed fuzzy control method together with the proposed control system has been shown effective and useful in not only achieving high-speed ACC control and low-speed stop & go control, but also fulfilling conventional cruise control.
VI. REFERENCES [1] F. Sanchez, M. Seguer, A. Freixa, P. Andreas, K. Sochaski and R. Holze, “From Adaptive Cruise Control to Active Safety Systems,” SAE Technical Paper, no. 2001-01-3245, 2001. [2] Adaptive Cruise Control, http://www.i-car.com. [3] J.E. Naranjo, C. Gonzalez, J. Reviejo, R. Garcia, and T. de Pedro, “Adaptive Fuzzy Control for Inter-Vehicle Gap Keeping,” IEEE Transactions on Intelligent Transportation Systems, vol. 4, no.3, pp. 132-142, Sept. 2003. [4] P. H. Shi, Design and Implementation of an FPGA-based Intelligent Cruise Control System, MS thesis, Department of Electrical and Control Engineering, National Chiao Tung University, 2005. [5] M. Person, F. Botling, E. Hesslow and R. Johnsson, “Stop & Go Controller for Adaptive Cruise Control,” Proc. of IEEE International Conference on Control Applications, Kohala Coast –Island of Hawaii, Hawaii, USA , pp. 1692-1697, Aug. 22-27, 1999. [6] D. Ward, T. Bertram and M. Hiller, “Vehicle Dynamics Simulation for the Development of an Extended Adaptive Cruise Control,” Proc. of the IEEE/ASME International Conference on Adavanced Intelligent Mechatronics, Atlanta,USA pp. 730-735, Sept. 19-23, 1999. [7] J. W. Chen, Fuzzy Neural Networks based Adaptive Cruise Control, MS thesis, Department of Electrical and Control Engineering, National Chiao Tung University, 2002.
[8] B. Riley, G. Kuo, B. Schwartz and J. Zumberge, “Development of a Controlled Braking Strategy For Vehicle Adaptive Cruise Control,” SAE Technical Paper, no. 2000-01-0109, 2000. [9] T. lijima, A. Higashimata, S. Tange, K. Mizoguchi, H, Kamiyama, K. Iwasaki and K. Egawa, “Development of an Adaptive Cruise Control System with Brake Actuation,” SAE Technical Paper, no. 2000-01-1353, 2000. [10] D. Littlejohn, T. Fornari, G. Kuo, B. Fulmmer, A. Mooradian,K. Shipp, J. Elliott and K. Lee, “Performance, Robustness, and Durability of an Automatic Brake System for Vehicle Adaptive Cruise Control,” SAE Technical Paper, no. 2004-01-0255, 2004. [11] W. Prestl, T. Sauer, J. Steinle and O. Tshchernoster, “The BMW Active Cruise Control ACC,” SAE Technical Paper, no. 2000-01-0344, 2000. [12] ISO 15622: Transport information and control systems-Adaptive Cruise Control systems-Performance requirements and test procedures, 2002.
[14] G. R. Widmann, M. K. Daniels, L. Hamilton, L. Humm, B. Riley, J K. Schiffmann, D E. Schnelker and W. H. Wishshon, “Comparison of Lidar- Based and Radar-Based Adaptive Cruise Control Systems,” SAE Technical Paper, no. 2000-01-0345, 2000. [15] S. Miyahara, “New Algorithm for multiple Object Detection in FM-CW Radar,” SAE Technical Paper, no. 2004-01-0177, 2004. [16] L. Hamilton, L. Humm, M. Daniels and H. Yen, “The role of Vision Sensors in Future Intelligent Vehicles,” SAE Technical Paper, no. 2001-01-2517, 2001. [17] R. Dixit, L. Rafaelli, “Radar Requirements and Architecture Trades for Automotive Applications,” IEEE MTT-S International Conference, vol. 3, pp. 1253 - 1256, Jun. 8-13, 1997.
Acknowledgements The authors sincerely acknowledge the technical support from the Automotive Research & Testing Center,
[13] SAE J2399: Adaptive Cruise Control (ACC) Operating Characteristics and User Interface, 2003.
Changhwa, Taiwan, the Republic of China.
TPS
Throttle Position Sensor
ABS Speed Sensor
Brake Position Sensor
Analog Signal
Frequency to Analog Converter
Ethernet
Data By Pass Box
Computer A
Imc Data Recorder
Optical Fiber
RS-232
Display for Driver
Analog Signal
Analog Signal
Lidar Sensor
Computer B
dSPACE Controller PWM Command
Vacuum in
Vacuum in
Connect to Air
Pulling Brake Pedal by Booster
Fig. 1.
Connect to Air
Pulling Gas Pedal by Booster
Overall system structure.
Steer rope Atmosphere
Piston
2.
Electromagn etic valve
3. 1.
Vacuum connect with engine air
Atmospher
GN
Fig. 2.
A
C
B
Power of 12V
Cross-sectional view of the vacuum booster.
Fig. 3. A recent picture of the experimental vehicle.
1. Relative Distance via Lidar (Dr) 2. Host Vehicle Speed from the ABS sensor (Vh) Set by Driver Desired Target Speed Vt
Check if any Vehicle Ahead Fail
True
Safety Distance Ds1 = 100m
Calculate Preceding vehicle’s speed Vp
Set by Driver Desired Time Gap tg
Calculate Safety Distance Ds1
Calculate Safety Distance Ds2
Find the Minimum of the Safety Distance Ds1 and the Safety Distance Ds2,i.e.,Ds=min(Ds1,
Ds2)
Calculation for the two inputs of Fuzzy Longitudinal Controller Relative speed (ΔV)
Distance error (Δd) Fuzzy Longitudinal Controller
u
Fig. 4. Architecture of the proposed fuzzy longitudinal controller.
Target speed (Vt) Time-gap (tg) Relative distance (Dr)
Calculations Distance error (Δd) of the distance error and the Relative speed (ΔV) relative speed
Fuzzy Longitudinal Controller
u>0
uthr
Acceleration Actuator
u
Automotive Research & Testing Center, Changhwa, Taiwan, R.O.C.
Department of Electrical Engineering, National Chung-Hsing University, Taichung, Taiwan
Abstract This paper presents a fuzzy longitudinal control system with car-following speed ranging from 0 to 120 km/h. This controller achieves the main functions of both adaptive cruise control (ACC) and Stop & Go control. For constructing the vehicular longitudinal control system, we use vacuum boosters to control the throttle and the braking pedal, thus circumventing the technical difficulties of using engine management system and anti-brake system. A fuzzy controller is synthesized by inputting the difference of the actual relative distance and the safe distance obtained from the preceding vehicle, and the relative speed, and then outputting the pulse-width-modulation (PWM) signal to control the output forces of the vacuum boosters. With the use of the high-level controller from dSPACE, the fuzzy control law is easily and rapidly implemented using Simulink for the experimental car, and the controller’s parameters can be changed and updated by analyzing the collected data from the relative distance from the lidar, the speed of the host vehicle, the opening of the throttle and the position of the braking pedal. Modified by a commercial SUV, the experimental car is equipped with a lidar on the bumper, a speed sensor mounted at the shaft of a wheel, and two sensors for measuring the opening of the throttle and the position of the braking pedal. For the sake of safe testing, experimental results are conducted by virtually simulating the various possible car-following conditions for the ACC and Stop & Go controllers, thereby obtaining virtually relative distance and speed to tune the controller’s parameters and make sure the safety of the controller. Afterwards, the realistic car following experiments are then conducted to confirm that the proposed fuzzy controller is capable of achieving the requirements of comfort and safety and giving a satisfactory performance at high and low speed conditions.. Keywords ： Adaptive cruse control, fuzzy control,
I.
INTRODUCTION
Currently ACC has been seen as a comfort system [1] for use to reduce driver fatigue in heavy traffic, but it is has the possibility to provide a lot of safety related feature in the near future and become an integral part of an intelligent transportation system (ITS). ACC is the longitudinal control for the vehicle dynamics; in the near future it combines the lateral control via active steering system and lane recognition systems or the intelligent transport infrastructure (e.g., high precision GPS), the dynasty of auto-driving will come true. ACC systems have been in market as an optional device for luxury vehicles. They have been developed for highway driving assistance, but could not handle urban traffic situations. One limitation of these commercial systems is that they control the speed of the car only at speeds above 40 km/h and they fail at lower speeds in heavy traffic. The failure may result from two key reasons. One is sensing difficulty in complex urban. The front vehicle situation of cut-in and lane change are more frequency, and the recognition pedestrian must be completely considered. The other is the frequent switching between acceleration and deceleration, and the smooth control for the driving stability and comfortable at low speeds is more difficult than at high speeds. The commercial systems [2] achieve vehicle longitudinal dynamics control via integration engine management system (EMS) and anti-lock braking system (ABS), thus their implementation do not need additional hardware. Considering the requirement of aftermarket and there are no independence for the two key systems at the local automotive manufacturers, we need to design a new type of by-pass actuator to control the vehicle longitudinal dynamics of acceleration and deceleration. In this paper, we desire to develop a longitudinal control method that could satisfy the driving condition between 0 to 120km/h, actually including the functions of ACC and Stop&Go control.
longitudinal control, stop & go. To achieve both functions of ACC and Stop & go, we
have to mount required sensors in the front of the car, measuring the preceding vehicle’s velocity and distance. The sensor could either be of optic or radar type, but the radar sensor is often preferred since it is much less influenced by the weather conditions than the optical sensor. The sensor information is then transmitted to the system’s electrical computer unit that controls the engine and the brake system. For safety reason, the car controllers only allow gentle acceleration and deceleration (under 0.3g) such that the driver should always be able to intervene if the system does not comply with the driver’s intentions. It is important to remember that the system is only a service to help the driver, not a replacement of the driver. Fuzzy logic is a powerful technique; in particular, it permits control without extensive knowledge of the equations of the plants or processes and it represents in a very effective way the human reasoning methods [3], [4]. The conventional PI controller was applied to the Stop&Go system [5]. This kind of controller [5] was divided into two separate parts: the outer control loop generates a desired acceleration that represents the driver behavior and the inner control loop controls the brake pressure and the throttle position that is dependent of the vehicle dynamics. Computer simulations are always helpful and useful in providing an assistance tool to facilitate the development of the ACC and the Stop&Go control, thus reducing development time and removing the risk on testing. The multi-body vehicle simulation tool Fasim_C++ provides an ideal environment to simulate vehicle behavior with ACC [6]. The longitudinal vehicle model has been used to accomplish ACC and Stop&Go control. The longitudinal controller is mainly composed of an engine, a torque converter, a transmission, a drive train, and a brake system. The simulation of the longitudinal controllers with fuzzy logic and neural networks based controller have been shown capable of giving good performance that provides the information of membership design [7]. Traditional cruise control is a universal system today, so acceleration control is not special technique by electronic throttle control. Deceleration Control for braking is typically performed using a “smart” booster, brake-by-wire, or an ABS modulator (hydraulic mechanization). Deceleration control for using a “smart ” booster is done by controlling the air flow valve of the booster, thereby regulating the vacuum inside the booster. Deceleration control using a brake-by-wire implementation has been performed by sending a brake command to the actuator at each wheel, which applies braking to the wheels. Deceleration control using the ABS modulator is implemented by regulation of wheel pressure through solenoid valves within the modulator [8], [9], [10]. ACC systems can provide assistance to the driver in a lot of situations, but never can relieve the drive’s
responsibility for vehicle guidance. The driver always has full responsibility for the driving task. For safety design, it is necessary to limit the systems deceleration and acceleration capability in order to avoid irritation of the driver and the surrounding traffic in case of inappropriate control reaction. In practical use a range od-2.0 m/s2 to +1 m/s2 was found to be a suitable compromise between customer benefit, convenience and safe requirement. It is necessary that the driver is allowed to override the system at any time by throttle or brake it without any conflicts with system functions [11], [12], [13]. Another critical point to a successful ACC system is the high performance sensor that has capability to detect the vehicle ahead of the host-vehicle, and measure the kinematic attributes of the target (e.g. distance, and relative velocity, etc.). For the need of the long sensing distance and reliability, lidar- and radar- based object detection technologies have been widely used. They need fast reaction to the dynamically changing forward traffic behavior (e.g. cut-in, lane changes, etc.) in the presence of various weather conditions (e.g. rain, fog, etc). Their requirement include the update time of less than 100 msec., the measurement range of relation speed from –150 km/h to 150 km/h, and the measurement range of the relative distance more than 100m [14], [15]. The advantage in CMOS imaging sensors has enabled low-cost and high-quality cameras that are making their way into future automobiles. In the future, vision sensors can be deployed in a car to improve robustness of the ACC system [16], [17]. The paper aims at developing a fuzzy longitudinal controller and investigating its feasibility and efficacy through several experiments. The contributions of the paper are fourfold. First, vehicular longitudinal control design is accomplished by using vacuum boosters to control the opening of the throttle and the position of the braking pedal. Worthy of mention is that this design does not involve with the complicated and sophisticated knowledge of the engine management system and the anti-brake system. Second, a single fuzzy controller is designed to carry out three control functions, including conventional cruise control, ACC and Stop&Go control. Third, the parameter tuning of the fuzzy controller is done by employing virtual car-following simulations. This kind of virtual experiment is safe and effective in tuning the controller’s parameters in an off-line manner. Finally, low-speed Stop & Go control can be easily accomplished by the proposed fuzzy control method. In addition, through experimental results, the proposed fuzzy method together with the developed control system has been shown effective and useful in not only achieving high-speed ACC control and low-speed stop & go control, but also fulfilling conventional cruise control. The remaining sections of the paper are outlined as follows. Section 2 details the novel control system architecture and its key components of the experimental platform. In Section 3, a fuzzy control approach is
briefly reviewed and a fuzzy longitudinal controller is presented in detail. Section 4 conducts several experiments to show the efficacy and merit of the proposed controller. Section 5 concludes the paper.
II. SYSTEM ARCHITECTURE AND DESCRIPTION This section aims at presenting new control architecture of the proposed fuzzy longitudinal control system with brake actuation, and describing the key components used for constructing such a system. This new design does not change or modify the original engine management system (EMS) and anti-brake system (ABS), but inserts additional components to achieve vehicular longitudinal control. Unlike the conventional adaptive cruise control systems, the proposed longitudinal control system uses vacuum boosters as main actuators for the throttle and the braking pedal, thus providing better controls to steer the vehicle; the kind of control system with the vacuum boosters is particularly useful in offering more aggressive control actions to brake the vehicle in case of emergent braking of the preceding car. In comparison with conventional actuators made by the direct-current motors, the proposed actuators are able to actuate the throttle and the braking pedal in both forward and backward directions, whereas the conventional actuators driven by the DC motors are only suitable for only one-direction driving. 2.1 Description of the Overall System Structure This subsection is devoted to describing the overall system structure of the proposed fuzzy longitudinal control system. Fig.1 shows the overall system architecture of the proposed control system. Four sensors, including the Lidar, the ABS speed sensor, the throttle position sensor, and brake position sensor, are used to provide all the necessary signals for the control system. The data recorder is responsible for collecting the signals from the four sensors. The data recorder is connected via Ethernet to the first computer, called Computer A, which monitors all the signals in real time. Worthy of mention is that all the signals must be connected to the data recorder For the purpose of real-time feedback control, the two signals obtained from the Lidar and the ABS speed sensor are directly inputted to the high-level controller, the MicroAutoBox 1507,made by dSPACE, thus generating the desired PWM commands to drive the two vacuum boosters.. The loop refresh rate (sampling) is performed once every 1/100 of a second. The second computer, named “Computer B”, is adopted to write and edit the control codes of the proposed control laws for the dSPACE high-level controller, tune the controller’s parameters, and acquire and display the real-time command information via the optical fiber network. The
display unit for driver attains the real-time command information and the current status of the experimental platform from Computer B via RS-232, and shows these messages to the driver. The main actuators are made by the two vacuum boosters, offering the driving and braking forces for the experimental platform. 2.2 Throttle and Braking Actuators This section uses the two vacuum boosters to design the braking pedal actuator and the gas pedal actuator. Fig.2 depicts the cross-sectional view of the vacuum booster. The pressure difference between the two sides of the piston produces a force to pull the gas or braking pedal through the steel rope. The pressure at the upper side of the piston is almost 1 atm because this air room is directly connected to the air. The pressure at the lower side of the piston can be regulated by exciting the three coils in the three electromagnetic valves. The electromagnetic valve B is used to adjust the opening of the valve by its PWM commands, thus reducing the pressure of the chamber inside the piston. Both two electromagnetic valves (A and C) are utilized to increase the pressure of the chamber inside the piston. To simply the control of the vacuum booster, we let the valve A be always closed and control the valve C by using PWM commands. For the sake of simple control, the two valves C and B are controlled by using the same PWM commands. Next, we move to the design of the braking pedal actuator. As shown in Fig.1, the vacuum booster is adopted to achieve the function by directly pulling the braking pedal. Without the pulling force, the braking pedal will be pushed back the original position by its internal spring. When the two forces are equal, the braking pedal will stay at an equilibrium position. Hence, the designed actuator will regulate the position of the barking at any desired location by applying appropriate PWM commands to the vacuum booster. The same design philosophy can be applied to the gas pedal actuator. Similar to the braking pedal actuators, the vacuum booster is directly used to pull the gas pedal, thereby changing the opening of the throttle via the steel rope. As shown in Fig.1, the gas pedal will recover to the original position by its internal spring while releasing the pulling force. Accordingly, the gas pedal actuator will be easily controlled by appropriate PWM commands. 2.3 Experimental Platform The experimental car used in our study was 1997 model-year HONDA SUV. This experimental platform was equipped with an automatic gearbox in order to simplify the gear change. With all the components mentioned before, the experimental car is equipped with the two computers, the data recorder, the dSPACE high-level controller, the display unit for driver, the four sensors and the two actuators, and the power supply.
Fig.3 presents the appearance of the experimental vehicle, where the lidar is mounted on the bumper of the experimental car.
III. FUZZY LONGITUDINAL CONTROLLER DESIGN This section presents a fuzzy longitudinal control method to achieve both high-speed adaptive cruise control and low-speed Stop & Go control. The conventional adaptive cruise control is usually concerned with the control of the host vehicle with speeds up to 40 km/h, in order to keep a safe distance away from the preceding vehicle. In contrast to adaptive cruise control, the stop & go control is often applied to steer the host car with speeds less than 40 km/h, dealing with the car-following problem with a safe distance away from the preceding vehicle. In the section, the proposed fuzzy control method is capable of addressing both two issues in a unified control framework. 3.1 Longitudinal Controller This subsection is aimed to describe the way of how to achieve the longitudinal control and to explain the proposed control logics to accomplish the conventional cruise and the adaptive cruise control. Under normal driver control, the acceleration and deceleration of the vehicle is maintained by use of the throttle alone. The throttle is opened to accelerate the vehicle, but closed to decelerate the vehicle due to engine braking effect. This implies that the longitudinal controller can decelerate the vehicle by making use of either the throttle or the normal braking. For braking, engine braking is used for the small deceleration whereas the normal braking is adopted for the larger decelerations. Although it is important to use the longitudinal controller to carry out precise speed tracking and car-following, the throttle control is still the main driver strategy under the consideration of the driver’s comfort and the fuel consumption. The normal braking control is utilized only in case of emergency, thus avoiding the frequent switching of the throttle control and the normal braking control. Although the cruise control and the so-called adaptive cruise control are done independently, this section proposes a longitudinal controller to accomplish precise speed tracking and car-following. Note that the proposed longitudinal controller works for the speeds ranging from the standstill condition to 120 km/h. Fig.4 describes the architecture of the proposed longitudinal controller. The relative distance via the lidar and the speed of the host car from its ABS speed sensor are considered as two inputs of the longitudinal controller. The relative distance information is taken to judge whether the preceding car appear in front of the preceding car. If there exists a preceding car, then the safe distance between the two cars must be evaluated by
taking into account the speed of the preceding car and the time gap set by the driver. The time gap is referred to the human reaction time from the moment that the driver is aware of danger to the moment that the driver takes actions to avoid the risk of danger. In other words, the time gap means that the minimum safe time to prevent the driver from dangerous conditions. In general, the time gap roughly ranges from one second to two seconds, according to the study of human agronomy. In this experiment we set the maximum of two seconds. In Fig.4, there are two other conditions to calculate the safety distances. The first case is that the safety distance is assumed to be one hundred meters if no preceding cars are found. The second case is that the safety distance B is regarded as the product of the desired time gap and the desired cruise speed specified by the driver. The final safety distance is the minimum of the sate distance A and the safety distance B. The proposed fuzzy longitudinal controller will be activated with the two inputs of the final safety distance and the related speed between the host car and the preceding car. Worthy of mention is that the proposed longitudinal controller will limit the speed of the host car to the desired speed if no preceding cars occur, thus preventing the host car from monotonically increasing its speed to impractical infinity. 3.2 Fuzzy Controller Design This subsection presents the proposed longitudinal control method. Fig.5 describes the block diagram of the fuzzy longitudinal control system. In Fig.5 the outputs of the longitudinal controller are the distance error and the relative speed between the host and preceding cars, which are the inputs of the fuzzy longitudinal controller. The relative speed Vr is calculated from the speed difference of the preceding and the host cars, i.e.,
ΔV = V p − V h
(1)
where Vp and Vh denote the speed of the preceding car and the speed of the host car. The distance error ed is defined by subtracting the relative distance from the safety distance, i.e.,
Δd = D s − D r
(2)
where Ds and Dr denote the safety distance and the relative distance, respectively. It is worthwhile to note that the relative distance is measured by the lidar and the safety distance is obtained from the product of the desired time gap and the speed of the preceding car plus the safety factor, that is ,
Ds = V p * t g + f
(3)
where t g and f denote the time gap and the safety factor, respectively. The safety factor means the offset distance to avoid car collision if the speed of the preceding vehicle becomes zero. In doing so, we set the
sufficiently safe distance of 10 meters in this experiment. After executing the fuzzy control rules specified by the designer, the fuzzy longitudinal controller will generate the actual control signal u. Note that the control signal u is in the interval [-1,1]. If the control signal u is positive, then the control u is transformed into a corresponding PWM command and the gas pedal actuator will be activated according the PWM command. Conversely, if the control signal u is negative, then the braking pedal actuator will provide the braking action, based on the braking PWM signal which is proportional to the absolute magnitude of the control u. Based on the PWM command, the gas or braking pedal actuator will produce a pull force to control the opening of the throttle and the position of the braking pedal, hereby accelerating or decelerating the host car and eventually obtaining the desired speed. In case of no preceding cars, the fuzzy longitudinal control will work in the cruise control mode. In the cruise control, the fuzzy longitudinal controller is used to achieve precise constant speed tracking and speed profile tracking. However, the physical quantities, including the safety distance DS , the relative speed ΔV and the distance error Δd , must be modified as follows.
DS = Vt × t g + f
(4)
ΔV = Vt − Vh
(5)
t
Δd = D s − ∫ ΔV × d τ 0
(6)
where Vt is the target speed specified by the user. Fig.6 depicts the fuzzy sets and the rule matrix of the fuzzy longitudinal controller. The membership functions of the input linguistic variables are triangular, and the output membership function of the control signal u is also triangular, thus simplifying the computations of the fuzzy controller. The two inputs are fuzzified into five linguistic variables, including NL (negative large) , NM (negative medium) , ZE (near zero), PM (positive medium) and PL (positive large), and so is the output u. However, the membership functions of the linguistic variables of the control u are different from those of the inputs. As shown in Fig.6; the linguistic variable ZE of the output significantly overlaps the linguistic variables NM and PM. Such a design eliminates frequent switching of the control signal from positive to negative, and causes the vehicle to move smoothly and comfortably, thereby avoiding abnormal fuel consumption. The fuzzy control rules are constructed based on the driver’s experience, thereby producing simple 25 control rules. These fuzzy driving rules are easily comprehended. For example, if the relative speed is NL, and the distance error is NL, then the control u is NL. This rule implies
that the host car is fast approaching the preceding one and the two cars are very close, then the large braking action should be taken to significantly reduce the speed of the host car in order to avoid car collision. Moreover, the rule matrix is symmetric with respect to the diagonal entries. Last but not least, the fuzzification ranges of the linguistic variables for the two inputs can be adjusted so as to achieve satisfactory accelerations and decelerations of the host car. The modification of the overall fuzzification ranges of the two inputs may also provide better drivers’ comfort. For example, if the fuzzification interval of the relative speed becomes narrow, then the accelerations and decelerations of the host car will turn out large, thereby causing the driver to feel discomfort. Conversely, if the interval of the relative speed becomes large, then the host car will have mild accelerations and decelerations and the driver will feel no discomfort. The situations also hold for the distance error, but the change of the fuzzification interval of the relative speed gives more effective tuning. Thus, the effective fuzzification intervals of both the distance error and the relative speed can be experimentally determined so that the desired accelerations and decelerations are achieved. Through our experiments, the useful fuzzification intervals for the distance error and the relative speed are selected as the interval [-30, 30] and the interval [-10,10], respectively.
IV. EXPERIMENTAL RESULTS AND DISCUSSION This section aims at conducting several experiments to examine the feasibility and effectiveness of the proposed longitudinal control system and fuzzy longitudinal control method. These experiments used the following settings: the desired time gap was two seconds and the safety factor was two meters. Before the performance of the overall system and the proposed control method were tested, it is necessary to investigate the input-output properties of the vacuum boosters and the characteristics of the Lidar. Several virtual simulations were employed to on-line tune the parameters of the fuzzy longitudinal controller in order to obtain satisfactory performance. Afterwards, the simple constant speed command tracking and speed profile tracking experiments of the fuzzy longitudinal controller were performed. Finally, the low-speed stop & go experiment was conducted in a realistic road environment. Fig. 7 and Fig. 8 respectively depict the performance of the proposed fuzzy longitudinal controller for conventional cruise control at 80 km/h and speed command profile tracking with real measurements from the Lidar. These experimental results indicate that the proposed fuzzy longitudinal controller is shown capable of achieving conventional cruise control and speed commands profile tracking. Worthy of mention is that the result in Fig. 7 (c) showed that the braking pedal was not activated for accomplishing the constant speed cruise
control, but in Fig. 8 (c) the throttle complemented the braking actuation for carrying out the constant command profile tracking. This fact indicates that the proposed controller behaved like a human driver.
systems are carried out using two independent controllers. The significant feature of the proposed method hinges on its capability to achieve ACC and stop & go control in a unified control framework.
Afterwards, this type of controller was applied to steer the host car in achieving the goal of adaptive cruise control. The stop & go experiment was setup as follows. The preceding car moved in the speed of 20 km/h and it was far from the host car. The host car moved forward with a speed of 50 km/h. In such a way, the relative distances between two cars were gradually reduced and the proposed fuzzy longitudinal controller mounted on the host car started to work. At the same time, the preceding care took actions of acceleration and deceleration, thus providing the time-varying speed profiles as shown in Fig.9(a). Fig.9(a) shows that the host car tracked the preceding car with small speed errors. Moreover, the result in Fig.9 (b) clearly indicates that the host car followed the preceding car at safe distances, and the result in Fig.9 (c) reveals that the time gaps between the two cars were greater than 3 seconds. As can be observed in Fig. 9 (d) and (e), the throttle worked with the braking pedal in a human driving manner, that is, the throttle complemented the braking actuation so that the proposed fuzzy longitudinal controller performed well in keeping safe inter-vehicle time gap.
Third, the parameters of the fuzzy controller are experimentally tuned by employing the virtual car-following simulations. These simulations take into account various real car-following conditions of the ACC and Stop & Go systems. These virtual simulations have been proven safe and effective in off-line tuning the controller’s parameters. The tuned parameters are used for real ACC and stop & go experiments.
V. CONCLUSIONS This paper has presented techniques to establish a new kind of adaptive cruise control system and to present a fuzzy control method. The main techniques developed in the paper include how to design vacuum boosters for actuating the throttle and the braking pedal, how to set up the lidar, sensors for measuring the angles of the throttle, the positions of the braking pedal and the speed of the host vehicle, how to synthesize the all-speed fuzzy controller, and how to proceed with real car-following experiments. The effectiveness and performance of the proposed system and method have been investigated in detail. The main results of the paper are summarized as follows. First, the vehicular longitudinal control design is accomplished by using feedback signals from Lidar and sensors for measuring the angles of the throttle, the positions of the braking pedal and the speed of the host vehicle, and by employing vacuum boosters to control the opening of the throttle and the position of the braking pedal. It is worthwhile to note that this new design does not change or modify the original engine management system (EMS) and anti-brake system (ABS), but inserts additional components to achieve vehicular longitudinal control. Second, a single fuzzy controller is synthesized to accomplish out conventional cruise control, ACC and Stop & Go control. The type of controller is a breakthrough because the current ACC and stop & go
Finally, low-speed Stop & Go control can be easily accomplished by the proposed fuzzy method. Conversely, conventional approaches are hard to achieve stop & go control in a smooth way of acceleration and deceleration. In addition, through experimental results, the proposed fuzzy control method together with the proposed control system has been shown effective and useful in not only achieving high-speed ACC control and low-speed stop & go control, but also fulfilling conventional cruise control.
VI. REFERENCES [1] F. Sanchez, M. Seguer, A. Freixa, P. Andreas, K. Sochaski and R. Holze, “From Adaptive Cruise Control to Active Safety Systems,” SAE Technical Paper, no. 2001-01-3245, 2001. [2] Adaptive Cruise Control, http://www.i-car.com. [3] J.E. Naranjo, C. Gonzalez, J. Reviejo, R. Garcia, and T. de Pedro, “Adaptive Fuzzy Control for Inter-Vehicle Gap Keeping,” IEEE Transactions on Intelligent Transportation Systems, vol. 4, no.3, pp. 132-142, Sept. 2003. [4] P. H. Shi, Design and Implementation of an FPGA-based Intelligent Cruise Control System, MS thesis, Department of Electrical and Control Engineering, National Chiao Tung University, 2005. [5] M. Person, F. Botling, E. Hesslow and R. Johnsson, “Stop & Go Controller for Adaptive Cruise Control,” Proc. of IEEE International Conference on Control Applications, Kohala Coast –Island of Hawaii, Hawaii, USA , pp. 1692-1697, Aug. 22-27, 1999. [6] D. Ward, T. Bertram and M. Hiller, “Vehicle Dynamics Simulation for the Development of an Extended Adaptive Cruise Control,” Proc. of the IEEE/ASME International Conference on Adavanced Intelligent Mechatronics, Atlanta,USA pp. 730-735, Sept. 19-23, 1999. [7] J. W. Chen, Fuzzy Neural Networks based Adaptive Cruise Control, MS thesis, Department of Electrical and Control Engineering, National Chiao Tung University, 2002.
[8] B. Riley, G. Kuo, B. Schwartz and J. Zumberge, “Development of a Controlled Braking Strategy For Vehicle Adaptive Cruise Control,” SAE Technical Paper, no. 2000-01-0109, 2000. [9] T. lijima, A. Higashimata, S. Tange, K. Mizoguchi, H, Kamiyama, K. Iwasaki and K. Egawa, “Development of an Adaptive Cruise Control System with Brake Actuation,” SAE Technical Paper, no. 2000-01-1353, 2000. [10] D. Littlejohn, T. Fornari, G. Kuo, B. Fulmmer, A. Mooradian,K. Shipp, J. Elliott and K. Lee, “Performance, Robustness, and Durability of an Automatic Brake System for Vehicle Adaptive Cruise Control,” SAE Technical Paper, no. 2004-01-0255, 2004. [11] W. Prestl, T. Sauer, J. Steinle and O. Tshchernoster, “The BMW Active Cruise Control ACC,” SAE Technical Paper, no. 2000-01-0344, 2000. [12] ISO 15622: Transport information and control systems-Adaptive Cruise Control systems-Performance requirements and test procedures, 2002.
[14] G. R. Widmann, M. K. Daniels, L. Hamilton, L. Humm, B. Riley, J K. Schiffmann, D E. Schnelker and W. H. Wishshon, “Comparison of Lidar- Based and Radar-Based Adaptive Cruise Control Systems,” SAE Technical Paper, no. 2000-01-0345, 2000. [15] S. Miyahara, “New Algorithm for multiple Object Detection in FM-CW Radar,” SAE Technical Paper, no. 2004-01-0177, 2004. [16] L. Hamilton, L. Humm, M. Daniels and H. Yen, “The role of Vision Sensors in Future Intelligent Vehicles,” SAE Technical Paper, no. 2001-01-2517, 2001. [17] R. Dixit, L. Rafaelli, “Radar Requirements and Architecture Trades for Automotive Applications,” IEEE MTT-S International Conference, vol. 3, pp. 1253 - 1256, Jun. 8-13, 1997.
Acknowledgements The authors sincerely acknowledge the technical support from the Automotive Research & Testing Center,
[13] SAE J2399: Adaptive Cruise Control (ACC) Operating Characteristics and User Interface, 2003.
Changhwa, Taiwan, the Republic of China.
TPS
Throttle Position Sensor
ABS Speed Sensor
Brake Position Sensor
Analog Signal
Frequency to Analog Converter
Ethernet
Data By Pass Box
Computer A
Imc Data Recorder
Optical Fiber
RS-232
Display for Driver
Analog Signal
Analog Signal
Lidar Sensor
Computer B
dSPACE Controller PWM Command
Vacuum in
Vacuum in
Connect to Air
Pulling Brake Pedal by Booster
Fig. 1.
Connect to Air
Pulling Gas Pedal by Booster
Overall system structure.
Steer rope Atmosphere
Piston
2.
Electromagn etic valve
3. 1.
Vacuum connect with engine air
Atmospher
GN
Fig. 2.
A
C
B
Power of 12V
Cross-sectional view of the vacuum booster.
Fig. 3. A recent picture of the experimental vehicle.
1. Relative Distance via Lidar (Dr) 2. Host Vehicle Speed from the ABS sensor (Vh) Set by Driver Desired Target Speed Vt
Check if any Vehicle Ahead Fail
True
Safety Distance Ds1 = 100m
Calculate Preceding vehicle’s speed Vp
Set by Driver Desired Time Gap tg
Calculate Safety Distance Ds1
Calculate Safety Distance Ds2
Find the Minimum of the Safety Distance Ds1 and the Safety Distance Ds2,i.e.,Ds=min(Ds1,
Ds2)
Calculation for the two inputs of Fuzzy Longitudinal Controller Relative speed (ΔV)
Distance error (Δd) Fuzzy Longitudinal Controller
u
Fig. 4. Architecture of the proposed fuzzy longitudinal controller.
Target speed (Vt) Time-gap (tg) Relative distance (Dr)
Calculations Distance error (Δd) of the distance error and the Relative speed (ΔV) relative speed
Fuzzy Longitudinal Controller
u>0
uthr
Acceleration Actuator
u
