Development of Multiple Sensor Fusion Experiments for Mechatronics Education

Transcription

1 Proc. Natl. Sci. Counc. ROC(D) Vol. 9, No., pp Development of Multiple Sensor Fusion Experiments for Mechatronics Education KAI-TAI SONG AND YUON-HAU CHEN Department of Electrical and Control Engineering National Chiao Tung University Hsinchu, Taiwan, R.O.C (Received August 8, 1998; Accepted May 4, 1999) ABSTRACT The development of effective training equipment for students in engineering colleges to acquaint them with sensory devices in mechatronics is presented in this paper. This training is focused on sensors, transducers, signal conditioning circuits, and microprocessor interfaces. Software as well as hardware techniques for designing a dynamic measurement system were emphasized in this study. In this paper, a developed integrated sensor module for mechatronics experiments is described. This module consists of incremental shaft encoders, an electronic compass and a vehicle platform. The module will be used as a teaching instrument in the course sensors and transducers for sophomore and junior students in colleges and universities. Applying the Kalman filtering theory, we successfully fused shaft encoders and electronic compass to obtain improved position and heading angle estimation of a test vehicle. Using this experimental module, students can be trained in the concept and effectiveness of multiple-sensor fusion. Key Words: electronic compass, incremental shaft encoder, sensor data fusion, mechatronics I. Introduction Mechatronics is a multi-disciplinary field, which involves mechanical engineering, electronics and electrical engineering and computer science. It includes system analysis and design, control techniques, sensors and actuators, microprocessors, interfacing, etc. It has been recognized that mechatronics is one of the most important areas for technology development, especially for modern production engineering. Progress in microelectronics, engineering materials and software design is very rapid nowadays. It is, therefore, very important to teach students in colleges of engineering such advanced concepts and techniques in order to meet the needs of the future labor market. Sensors and sensing techniques are required in almost every mechatronic system. Due to the rapid evolution of materials and microelectronics, new types of sensors and transducers are introduced to the market almost every day. This in turn improves the performance of various mechatronic products. As for the training of engineering students in this area, it is well recognized that not only the theoretical principles about sensors should be taught, but also laboratory practice is essential for effective training. It is, therefore, desirable to develop effective and practical training equipment for students in colleges of engineering to acquaint them with sensor devices in the field of mechatronics. Techniques of system integration such as sensors, transducers, signal conditioning circuits, microprocessor interfaces should be taught to students exploiting various training equipment. Furthermore, it is important to teach the students both hardware and software techniques for developing dynamic measurement systems. Integrating multiple sensors into the process can provide more reliable sensing information. An effective method for sensor data integration is the basic requirement for a distributed sensory system. This paper presents the design and implementation of novel and advanced training equipment in the area of multi-sensor data fusion. The design goal was to investigate the problem of posture estimation of an autonomous vehicle. It is well recognized that posture estimation is extremely important in mobile robot navigation. Conventional methods, such as dead 56

2 Multiple Sensor Fusion Experiments reckoning, which have been widely used for wheeled mobile robots, utilize pulses from shaft encoders to calculate the robot s current location. The dead-reckoning method is simple and effective. However, the robot may fail to keep track of its desired location over long distances. This is mainly because in this method, the wheels are subject to major cumulated errors caused by wheel slippage. Several approaches have been proposed in the past few years to eliminate the drawback of the deadreckoning method in position estimation. Kim and Seong (Kim & Seong, 1996) used an encoded magnetic compass to compensate for the abnormal orientation drift caused by wheel slippage, thereby resulting in robot position recovery. The disadvantage of the method is that the magnetic compass does not function well at a place where the magnetic field varies from position to position. Another method based on active beacon positioning, is called the 3D location technique (Flgueroa & Mahajan 1994). In this method, a transmitter is mounted on the robot as a beacon, and several receivers are installed at pre-specified locations with respect to a reference coordinate system. Absolute locations can be obtained but the receivers are difficult to implement. Borenstein and Feng developed a method, termed the University of Michigan Benchmark (UMBmark), for quantitatively measuring systematic odometry errors and, to a limited degree, nonsystematic errors (Borenstein & Feng 1996). The calculation of all the terms of odometry errors is timeconsuming, so this method lacks flexibility. This topic is very interesting and suitable for students involved in a mechatronics program. Through practical experiments on robot localization, students will get acquainted with useful position sensors. Since several sensors such as shaft encoders and electronic compass can be utilized simultaneously on the mobile cart, advanced training on multiple sensor fusion can also be realized employing the equipment. A small mobile cart was deigned and constructed in this study to serve as an experimental platform. To improve confidence in and accuracy of mobile robot localization, Kalman filtering is adopted for fusing sensor from an electronic compass and dead-reckoning estimates employing shaft encoders. The electronic compass is added to overcome the wheel slippage problem. The role of the sensor data fusion system is to combine information from multiple, potentially disparate, information sources. The system has the potential to be more robust with respect to poor sensor performance or sensor failure in a multi-sensor system. This is due to the fact that other sensors will still be capable of providing information about the states. The remaining part of this paper is organized as follows. Section II describes the methodology of vehicle posture estimation based on shaft encoders and an electronic compass. Formulation of the Kalman filter to fuse sensor information from encoders and electronic compass is introduced. In Section III, three experiments are described and results given to verify the performance of the proposed system. Section IV concludes this paper. II. Methodology This section describes the theoretical background of the developed multi-sensor system for vehicle posture estimation. The fundamental idea of posture estimation is based on the measurement of wheel shaft rotation. Two shaft encoders are utilized to obtain the rotational displacements. Formulas (1)~(5) hold true when wheel revolutions can be translated accurately into linear displacement relative to the floor. One can calculate the integration of incremental motion information over time using the information from shaft encoders: ds r (k) and ds l (k). dθ e (k)= 1 w (ds r(k) ds l (k)) (1) ds(k)= 1 (ds r(k)+ds l (k)) () x(k)=x(k 1)+ds(k) cos( θ e (k)+θ e (k 1) ) (3) y(k)=y(k 1)+ds(k) sin( θ e (k)+θ e (k 1) ) (4) θ e (k)=θ e (k 1)+dθ e (k) (5) where x(k),y(k) : vehicle position at the kth instant; w : distance between two drive wheels; dθ e (k),ds(k) : variations of vehicle heading angle and position between the kth and (k 1)th sample instant; ds r (k),ds l (k) : displacement of left and right wheels between the kth and (k 1) th sample instant. The above described vehicle posture estimation method has been widely used in mobile robot selflocalization. But since the encoder measurement is based on the wheel shaft, it leads inevitably to unbounded accumulation of errors if wheel slippage occurs. Specifically, orientation errors will cause large lateral position errors, which will increase proportionally with the distance traveled by the robot. In this study, an electronic compass was applied to measure 57

3 K.T. Song and Y.H. Chen 1. Shaft Encoders Fig. 1. The experimental setup. Optical incremental encoders are widely used in industries for shaft angular displacement measurement. Normally the encoder is mounted on the motor shaft, and one has to power the motor in order to acquire and observe the encoder signals. In this study, we used an experimental apparatus which could be tested with encoders easily. As described in the previous paragraph, one can estimate the position and heading angles of the vehicle platform using the feedback information from the two encoders on the left and right wheels of the vehicle platform. Figure 4 shows the encoders used in the experiments. The shaft encoders used in this work were LBJ series produced by the Sumtak Company. They can generate 500 pulses per cycle and 000 quadratural counts. The output of each shaft encoder consists of two phases: phase A and phase B with 90 degrees phase difference. One can identify the rotary direction of the shaft by observing the phase difference, which is shown in Fig. 5. The encoder is up counting while Fig.. The test vehicle. the orientation of the vehicle and fuse with the heading angle estimated by shaft encoders to overcome the problem of accumulation of errors. A simple experimental vehicle was constructed in this study. It was not a practical useful vehicle, but was designed for future application in students lab work. The system architecture of the vehicle platform is shown in Fig. 1. It consists of three main parts: (1) a three-wheeled platform; () an I/O interface card; (3) a host computer. Figure illustrates the physical structure of the testing vehicle. Both of the front wheels were equipped with shaft encoders. A free caster was put on the rear wheels for balance. The electronic compass was mounted on top of the vehicle platform. The wheels were not powered. The platform was operated manually. A PC add-on card was developed to interface the sensory data with the computer (Fig. 3). This card contains two parallel-interface ICs (855A) for interfacing to the computer and two HCTL- 00s for decoding and counting the pulses from the shaft encoders. The signal from the electronic compass, after serial to parallel conversion, was sent to the host computer by using two I/O ports on the interface card. A user-interface in the host computer was designed to display the processed sensory information. Fig. 3. The I/O interface card. Fig. 4. Shaft encoders were located separately on the left and right wheels. 58

4 Multiple Sensor Fusion Experiments Fig. 5..Output signals of the shaft encoder. the state is and down counting while the state is Electronic Compass A low cost, relative low precision electronic compass was utilized in this study. It can measure the heading angle of a moving object. The advantage of this approach is that the angular position obtained is relative to the earth. Therefore, it is not influenced by the wheel slippage that occurs on the test vehicle. The compass module Vector X was produced by Precision Navigation INC. It is a two axis compass module suitable for OEM applications. Using PNI s patented magneto-inductive magnetometer technology, Vector X provides acceptable accuracy, low power, and low cost in a small package. The module delivers -degree accuracy with 1-degree resolution. The sampling rate is five Hz. The main applications of this heading angular measurement module include a GPS positioning system, antenna angular indication and an automatic navigation system(precision Navigation. INC. 1996). An interface electronic circuit has to be designed in order to read the heading data from the module. The vector X provides two forms of digital output data, BCD codes and Binary codes. It can operate in three modes: master mode, slave mode, and raw mode. The raw mode delivers X-axis and Y-axis raw data for magnetic scale. In the slave mode, the host processor must generate the data clock; and the vector X outputs a single heading data when polled each time by the host. This is more complicated to interface and not efficient in our application. We utilized the master mode in this work. In master mode, the Vector X clocks out the data using an internally generated clock. The timing diagram of the signals generated in master mode is shown in Fig. 6. The M/S pin must be tied low to enable master mode. The P/C pin is used to request data from the Vector X, and the EOC is used to frame the data. The SS pin and SCLK pin are outputs in this mode; the SCLK pin outputs 16 clock pulses of 4K Hz for each conversion. The first six bits output by the SDO pin will always be zero, in either binary or BCD format; and the first significant bit (FSB) after the first six bits will always be a zero in binary mode. This enables the Vector X to load data into devices such as serial to parallel shift registers without the intervention of the host processor. Fig. 7 shows a block diagram of the conversion from serial data to parallel format accepted by the computer. We designed this circuit using the design tool MAX+PLUS II from ALTERA. The interface circuit was realized in an integrated circuit MAX7000. This is a complex programmable logic device (CPLD), which can be reprogrammed by the user. Hence, it is easy to redesign and modify the circuits. Another advantage of adopting this device is the size of circuit board was reduced. Figure 8 shows the module of the electronic compass and the CPLD chip.. Sensor Data Fusion The Kalman filter is a recursive algorithm used to optimize system states based on the observations which may not be very accurate (Haykin 1996). Assume that one has data from two different sensors (x a and x b ), and that one gives each of them a weight Fig. 6. Data clock timing for the Vector X in master mode [4]. Fig. 7. Block diagram of the digital compass module. 59

5 K.T. Song and Y.H. Chen equation: where θ fuse = θ e + σ e σ c + σ e (θ c θ e ) (9) Fig. 8. Vector X electronic compass and interfacing circiuts. (w a and w b ); then, the optimal estimation can be expressed as follows: x=w a x a +W b x b (6) θ fuse :the heading angle estimation after Kalman filter fusion; θ e :the heading angle estimation obtained from the shaft encoders; θ c :the heading angle estimation obtained from the electronic compass; σ e :variance of shaft encoders estimation; :variance of electronic compass estimation; σ c The resulting estimation based on data fusion, θ fuse, can then be substituted into (5) to reduce the accumulation of errors which occur in the process of vehicle pose estimation. Figure 9 depicts the process of sensor data fusion. where W a = w a W, W b = w b W W=w a +w b. If all the measurement errors can be modeled by a Gaussian distribution, then the weights are inversely proportional to the variance of the estimates. Suppose that we obtain two estimates x 1, x and the corresponding variance σ 1, σ ; the optimal estimate X can be expressed as follows: σ X =[ σ 1 + σ ] x 1 +[ σ 1 σ 1 + σ ] x (7) The variance of X, σ is determined by the following expression: 1 σ = 1 σ σ. Equation (7) can be simplified as follows: σ X = x +[ σ 1 + σ ](x 1 x ) = x + K(x 1 x ), (8) σ where K= σ 1 + σ. One can fuse information from the shaft encoders and electronic compass by means of the following Fig. 9. Flowchart of sensor data fusion. 60

6 Multiple Sensor Fusion Experiments Table 1. Calibration Results of Vector X Actual heading Estimated angle Estimated angle Error angle before calibration after calibration reduction based on shaft encoders. Figure11 shows the recorded experimental results. The estimation error in the X- axis direction was.4958 cm, and that in the Y-axis direction was cm. In the second stage of the experiment, we placed the vehicle platform at the predefined reference position and rotated the vehicle on the spot 180 clockwise and counterclockwise. After testing 50 times, the resulting estimation angles were recorded and shown in the following four figures. Figure1 depicts the estimated heading angle after rotation 90 clockwise with a variance of Figure13 shows the result Fig. 10. Vector X calibration experiment. III. Experimental Results Three experiments were carried out and are presented in this section to demonstrate the feasibility, accuracy and performance of the proposed localization system. These experiments were also developed for the students in the mechatronics lab. The first experiment focused on the observation of the orientation accuracy of the electronic compass after calibration. The second one focused on investigating how to estimate the position and heading angle of the vehicle platform using feedback information from the shaft encoders. We experimented with the sensor fusion method using compass and shaft incremental encoders in the third experiment. The results will be compared and discussed. The first experiment was performed to observe the accuracy of the electronic compass after calibration. We put the digital compass on a 360-degree rotary table to calibrate the compass as shown in Fig. 10. Table 1 presents the experimental results after calibration. The purpose of the second experiment was to estimate the position and heading angle of the vehicle platform while employing shaft encoders. In the first stage of this experiment, we guided the vehicle along a straight line for 50 cm. The trajectory was estimated Fig. 11. Experimental result of position estimation based on the shaft encoders. Fig. 1. The estimated heading angle based on shaft encoders after rotation clockwise with a variance of

7 K.T. Song and Y.H. Chen Fig. 13. The estimated heading angle based on shaft encoders after rotation 180 clockwise with a variance of Fig. 15. The estimated heading angle based on shaft encoders after rotation 180 counterclockwise with a variance of Fig. 14. The estimated heading angle based on shaft encoders after rotation 90 counterclockwise with a variance of after rotation 180 clockwise with a variance of Figure 14 depicts the estimated heading angle after rotation 90 counterclockwise with a variance of , and Fig. 15. shows the result after rotation 180 counterclockwise with a variance of In the third experiment, sensor data from the electronic compass and shaft encoders were fused for Fig. 16. Comparison of position estimation results with and without sensor data fusion. accurate position estimation. In this experiment, we pushed the vehicle along a predefined trajectory and recorded the fused positions. The trajectory consists of a horizontal straight line, a sharp turn and a vertical 6

8 Multiple Sensor Fusion Experiments Fig. 17. Comparison of heading angle estimation results with and without sensor data fusion. straight-line part. Figure 16 presents the experimental results. In the figure, two recorded trajectories are shown; one is based only on shaft encoders, and the other is based on sensor data fusion. The average error of trajectory estimation based on shaft encoders is cm in the X-axis direction and cm in the Y-axis direction. The average error of trajectory estimation based on sensor data fusion is cm in the X-axis direction and cm in the Y-axis direction. Figure 17 presents the heading angles of these two approaches. We see that the estimation with sensor fusion has improved performance. IV. Concluding Remarks It has been widely recognized that sensors and measurement techniques are essential for industrial control systems. The accuracy and reliability of the sensing and measuring process directly affect automatic control quality. It is clear that sensors and sensing technology play an important role in industry automation. The training of college students to familiarize them with sensor applications is important to upgrade domestic industries. On the other hand, due to the rapid development of sensors and sensing technology in recent years, it has become more important to provide new equipment and training mate- rials to improve the quality of education. In this study, a sensor experiment prototype has been developed for use in a mechatronic course. The experiment platform integrates two kinds of position sensors: (1) Vector X magnetic sensor for angular displacement measurement; and () optical shaft encoders for position and orientation measurement. The concept and techniques of multi-sensor fusion can be taught and conveyed to the students through this experimental setup. We have developed a sensor training module which focuses on three topics: (1) absolute heading angle measurement of an object using an electronic compass; () vehicle posture estimation based on incremental shaft encoders; (3) reduction of accumulated errors using sensor fusion which employs electronic compass and shaft encoders. Different sensing devices can acquire position and orientation information in different working situations. Since each sensor has its own limitations, techniques for fusing together several kinds of sensing information have become important (Iyengar & Prasad 1995). Experimental results confirm that our system based on sensor data fusion improves the accuracy and confidence of vehicle posture estimation. Meanwhile, the experimental platform and materials can provide college students with practical and efficient training on this subject. Acknowledgements This study was supported by the National Science Council of the Republic of China under grant NSC85-51-S EE. The authors acknowledge CIC, National Science Counsil of the ROC and ALTERA for providing the MAX+PLUS II software. References Borenstein, J., & Feng, L. (1996). Measurement and Correction of Systematic Odometry Errors in Mobile Robot. IEEE Trans. on Rob. and Autom., 1(6), Flgueroa, F., & Mahajan, A. (1994). A Robust Navigation System for Autonomous Vehicles Using Ultrasonics. Control Engineering Practice, (1), Haykin, Simon (1996). Adaptive Filter Theory (3rd ed.). Englewood Cliffs NJ: Prentice Hall, Inc. Iyengar, S.S., & Prasad, L. (1995). HLA MIN, Advances in Distributed Sensor Technology. Englewood Cliffs NJ: Prentice Hall, Inc. Kim, J.H., & Seong, P.H. (1996). Experiments on Orientation Recovery and Steering of Autonomous Mobile Robot Using Encoded Magnetic Compass Disc. IEEE Trans. on Instrum. And Meas. 45(1), Precision Navigation. INC. (1996). Vector X Compass Module, Application Notes, Version 1.03, January. 63

9 K.T. Song and Y.H. íê~åëçìåéêë!"#$ %&'()*+,-. / !"#$%&'()*+,-./0134^!"#$%&' ()*+, _!"# $%! `!"#$%&'()*+,-./01$%3 &45%67$89:;<= 64