Está en la página 1de 11

3402

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 6, NOVEMBER 2008

Civilian Vehicle Navigation: Required Alignment of the Inertial Sensors for Acceptable Navigation Accuracies
Zainab F. Syed, Priyanka Aggarwal, Xiaoji Niu, and Naser El-Sheimy
AbstractA vital necessity for any kind of inertial navigation system (INS) is the alignment of its axis with the vehicle body frame (VBF). Civilian vehicle navigation has strict requirements with respect to cost, size, reliability, and ease of implementation of the system. Microelectromechanical system (MEMS) inertial sensors have satised the cost and size requirements for civilian vehicle navigation; however, reliability and ease of implementation of these low-cost and miniaturized navigation systems are still parts of major research and investigation. This paper focuses on an important aspect of the ease of implementation for inertial sensors. From a civilian user perspective, accurately aligning the inertial system with respect to the vehicle, before every use, is not a desirable quality for a portable navigation system. In addition, it is not realistic to assume that even a careful user can achieve good alignment accuracy of the system. The purpose of this paper is to investigate the effects of misalignment errors that will produce errors in initial alignment and affect the navigation accuracy for two different inertial systems. The inertial systems are classied according to the number of sensors used in the system. The rst system consists of three gyros and three accelerometers [full inertial measurement unit (IMU)], whereas the second system only has one gyro and two horizontal accelerometers (partial IMU). Index TermsAlignment, Global Positioning System (GPS) and inertial navigation system (INS) integration, microelectromechanical systems (MEMS) inertial sensors, portable navigation system.

I. I NTRODUCTION HE GLOBAL Positioning System (GPS) provides accurate long-term position and velocity information at a relatively low-frequency data rate when a direct line of sight of four or more satellites exists. On the other hand, microelectromechanical systems (MEMS)-based inertial measurement units
Manuscript received June 21, 2007; revised December 7, 2007, March 6, 2008, and March 11, 2008. First published March 31, 2008; current version published November 12, 2008. The work of N. El-Sheimy was supported in part by the Natural Science and Engineering Research Council of Canada (NSERC) and in part by Geomatics for Informed Decisions (GEOIDE), Network Centers of Excellence (NCE). The review of this paper was coordinated by Prof. A. Tonello. Z. F. Syed is with the University of Calgary, Calgary, AB T2N 1N4, Canada, and also with Novatel Inc., Calgary, AB T2E 8S5, Canada (e-mail: zfsyed@ucalgary.ca). P. Aggarwal is with the University of Calgary, Calgary, AB T2N 1N4, Canada, and also with AeroMechanical Systems, Calgary, AB T2E 6Y7, Canada (e-mail: aggarwap@ucalgary.ca). X. Niu was with the University of Calgary, Calgary, AB T2N 1N4, Canada. He is now with SiRF Technology, Inc., Pudong, Shanghai 201206, China (e-mail: xjniu@sirf.com). N. El-Sheimy is with the University of Calgary, Calgary, AB T2N 1N4, Canada (e-mail: naser@geomatics.ucalgary.ca). Digital Object Identier 10.1109/TVT.2008.921616

(IMUs) are small and self-contained systems that provide navigation data at high rates [1]. However, for the MEMS IMU system, the data accuracy rapidly degrades with time due to the effect of various factors such as sensor biases, position drifts, and noises [2], [3]. Hence, when GPS and IMU are integrated, an accurate, drift-free, and high-resolution navigation system can be achieved for most outdoor situations. The integrated system can effectively deal with short GPS signal outages [4]. Inertial systems require alignment, whereby the orientation of the axes of an inertial system, which should follow the axes of the vehicle frame as shown in Fig. 1(a), is determined with respect to a reference frame such as local level frame (LLF), which is also known as navigation frame (n). The alignment can be achieved by leveling and gyro-compassing, in which the orientation of the system is estimated by equating the specic force from accelerometers and the Earths rate from gyroscopes to their expected values. For MEMS sensors, the strong gravity signals from accelerometers can be used to determine the angular displacements in the roll and pitch of the inertial system with respect to the reference frame. Here, it is important to note that the inertial system is aligned with the vehicle body frame (VBF), i.e., the rst set of accelerometer and gyro is pointing in the forward/backward direction, the second set associates with the sides of the vehicle, and, nally, the third set is in the vertical direction. Since the Earths rate cannot be observed due to the high noise of MEMS inertial sensors, therefore, a technique known as velocity matching is used to determine the heading of the vehicle. When the vehicle moves, the GPS velocity and the inertial system-derived velocities are compared and used to determine the heading of the vehicle. Once the inertial system is roughly aligned by the mentioned leveling and gyrocompassing method (or the velocity matching), ne tuning of the angular displacement vector in a Kalman lter (KF) further reduces any remaining errors. The alignment provides the direction cosine matrix (DCM) for the system that is used to convert the accelerometer-measured specic forces into the desired navigation frame [5]. As mentioned earlier, the rst requirement in using inertial systems in vehicle navigation is to align the axes of the inertial system with the axes of the VBF. If this step is not properly done, the DCM from the alignment process will be erroneous. The misalignment issues in this paper deal with the axes mismatch between the VBF and the IMU frame, as shown in Fig. 1(b).

0018-9545/$25.00 2008 IEEE

SYED et al.: CIVILIAN VEHICLE NAVIGATION: REQUIRED ALIGNMENT OF THE INERTIAL SENSORS

3403

Fig. 1.

Schematics of the misalignment of IMU with respect to VBF. (a) Well-aligned IMU with vehicle frame. (b) Misaligned IMU and vehicle axes.

Navigation accuracy can be improved by applying the nonholonomic constraint to IMU/GPS integration [7]. The nonholonomic system restricts the admissible directions of motion, e.g., a land vehicle can only move forward or backward in a direction perpendicular to the orientation of its rear wheel axle. Using these constraints, the roll and pitch of the vehicle can also be computed directly from IMU measurement when the forward acceleration is zero. However, if the forward acceleration is not zero, the differential equation for the forward velocity can directly be obtained as explained in [8]. In an ideal case, three accelerometers and three gyroscopes should be part of the IMU system to faithfully record the 3-D motions of the vehicle. Gyroscopes are usually more expensive units than accelerometers. For land vehicular use, the most important gyroscope is in the vertical axis that senses the heading of the vehicle. As a result, the recent direction of research is toward exploring minimum gyroscope usage in the IMU [6]. The consideration of such a system with a minimum number of sensors will require even more accurate alignment, particularly in the roll and pitch direction. Any angular displacement of the system will not be able to resolve the correct heading due to the lack of information available. Therefore, to design a costeffective portable navigation system, the following points are important and are answered in this paper. 1) What kind of sensor/vehicle alignment accuracy is required to obtain meaningful navigation results? 2) What is the required accuracy for the sensor-to-vehicle alignment if the conguration of the inertial sensors is incomplete in the IMU, i.e., less than six sensors? 3) Nonholonomic constraints can improve the navigation results. What kind of alignment accuracy is required for these constraints to benet the overall results? To summarize, the purpose of this paper is to estimate the effects of misalignment errors between the IMU and the vehicle [Fig. 1(b)]. This misalignment will produce errors in the initial alignment of the system and, thus, result in an erroneous navigation solution. Full and partial IMUs are studied,

and allowable misalignment limits for a meaningful navigation solution are deduced. II. B ACKGROUND A. Recent Technology and Need for Inertial Sensors The current navigation systems for land vehicle navigation include Tomtom Go [9], Sony NAV [10], Garmin [11], and Navman [12]. These systems are solely based on the usage of the GPS receiver to estimate the vehicles absolute location and digital maps to display the locations to the users. In addition to providing accurate vehicle location, these systems also offer a variety of route calculation and search algorithms for well-organized trip planning. GPS receivers can only estimate positions when there is a direct line of sight to four or more satellites. An obvious problem with any available system is the loss of satellite signals if the vehicle enters an urban canyon, forest canopy, or a tunnel. In this situation, the system ceases to work as there is no positioning information available to display. As a result, research has been targeted toward the integration of GPS with inertial navigation systems (INSs) to bridge the GPS signal gaps [13]; however, this type of sensor integration still needs to make its way into civilian navigation. The fundamental approach to integrate INS and GPS data is through the Extended Kalman Filter (EKF). B. EKF The Kalman Filter is a set of mathematical equations that provide an efcient computational mean to estimate the state of a linear process by minimizing the mean of the squared errors. However, if the process is nonlinear, the usual method is to linearize the process about the lters estimated trajectory, and it is termed EKF [16]. The EKF assumes that the system state vector xk evolves with time as xk+1 = k xk + wk (1)

3404

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 6, NOVEMBER 2008

where wk is the white noise sequence with covariances Qk , and k is the state transition matrix I + F t [7]. The state vector consists of position, velocity, attitude, and sensor bias errors as follows: xk = 3,1 rn 3,1 vn n 3,1 b6,1
T

importance when nonholonomic constraints are applied. The VBF velocity can be computed as n v b = Cb
T

vn

(6)

(2)

where 3,1 = [rx ry rz ]T are position errors; rn 3,1 = [vx vy vz ]T are velocity errors; vn n are errors in attitude. 3,1 = [x y z ]T In addition, b6,1 is the vector for three accelerometers and three gyroscope sensor biases. Initially, the covariance matrix corresponding to the above states (Pk ) will consist of the diagonal elements representing the variance of each state. The computed position, velocity, and attitude errors along with the sensor biases will be fed back to the control error growth for sensor-only measurements. The measurement vector Zk is given by Zk = Hk xk + vk (3)

n where n is the navigation frame, and (Cb )T is the rotation matrix from the navigation frame to the VBF. Only considering the rst-order error terms, the corresponding error equation will be
b v b Cn v n Cn (v n )n = b

(7)

where v b v n n

b b b [vx vy vz ] ; residual errors in the velocity vector; residual errors in the DCM that is used to rotate the navigation frame to VBF. The measurement equation can be constructed from the second and third rows of (7) as b vy b vz

where vk is the white noise sequence with covariances Rk , and Hk is the measurement matrix that relates the measurements with the states when there is no noise. The measurement vector and measurement matrix [7] will be of the form (M + h)(INS GPS ) (N + h) cos (INS GPS ) Zk = (4a) hINS hGPS vn vn INS GPS (M + h) 0 0 0 (N + h) cos 0 03,3 03,9 Hk = (4b) 0 0 1 I3,3 03,9 03,3 where is the longitude of the vehicle, GPS refers to measurements from GPS, and INS shows the navigation measurements using mechanization equations [1] on the inertial sensor signals. C. Nonholonomic Constraints Nonholonomic constraints can be used to further improve the navigation parameter accuracy, particularly when GPS outages occur. Nonholonomic constraints [6], [7] use the fact that a land vehicle cannot move sideways or vertically, and hence, these two velocity components should be zero as
b vy 0 b vz 0

Zk = Hk =

(8) K1,3 L1,3 vD C22 +vE C32 vD C23 +vE C33 vD C12 vN C32 vD C13 vN C33 vE C12 +vN C22 vE C13 +vN C23 01,6 01,6 (9)

01,3 01,3

where the rst column of the H matrix corresponds to the position errors in the state vector, the second column corresponds to the velocity errors, columns three to ve correspond to the angular displacements, and the last column takes care of the sensor errors K1,3 = [C12 L1,3 = [C13 C22 C23 C32 ] C33 ].

b Cxy s are elements of the DCM Cn . The estimated errors are then fed back to correct the IMU mechanization solution [1]. It is evident from (5) that correct knowledge of the n frame and VBF is the rst requirement for the proper application of the nonholonomic constraints. This is only possible when the sensor frame and VBF are the same as that given in Fig. 1(b).

III. M ETHODOLOGY A. Simulation of Misalignment

(5)

where b VBF; y sideways component of the velocity; z vertical component of the velocity. As evident from (5), velocity should be converted into VBF to apply the nonholonomic constraints. As a consequence, the alignment of the inertial system with the VBF is of prime

To study the effect of misalignment, a Matlab script that converts the well-aligned inertial signals from full IMU to the specied misaligned signals was developed. The desired misalignment angles are rst converted into DCM and then multiplied with the well-aligned inertial signals. 1) Calculating the DCM matrix: Given in the following are the sequences of rotations from the VBF to the IMU frame to obtain the DCM matrix when desired misalignment angles are introduced:

SYED et al.: CIVILIAN VEHICLE NAVIGATION: REQUIRED ALIGNMENT OF THE INERTIAL SENSORS

3405

Fig. 2. (a) Trajectory for data set 1 with eight simulated GPS outages (some outages are overlapped). (b). Trajectory for data set 2 with ve simulated GPS outages.

a) a rotation about the z-axis through the heading misalignment angle (); b) a rotation about the y-axis through the pitch misalignment angle (); c) a rotation about the x-axis through the roll misalignment angle (); 1 0 0 cos 0 sin DCM = 0 cos sin 0 1 0 0 sin cos sin 0 cos cos sin 0 sin cos 0 . (10) 0 0 1 2) Multiplying DCM with the IMU signals: Next, the DCM is multiplied with the well-aligned IMU data to obtain misaligned simulated signal values as xaligned xmisaligned ymisaligned = DCM yaligned . (11) zmisaligned zaligned In this paper, both accelerometer and gyro data are misaligned using the preceding formula. This will ensure that the analysis is unbiased and only represents the simulated misalignment. The range of simulated misalignments for the attitude varied from 45 to 45 . B. Test Trajectories and Simulated GPS Outages Data from two test trajectories are being studied in this paper. These were collected using a 2003 Dodge Grand Caravan test vehicle. The Analog Devices Inc. MEMS Sensor Triad (ADI) data were integrated with single point positioning (SPP) data from Novatels OEM4 receiver. The ADI unit was designed by members of the MMSS research group at the University of Calgary. The OEM4 GPS antenna was mounted on the roof of the test vehicle, and the ADI unit was installed inside the van. The rst trajectory started from a medium GPS signal quality, where the number of observed satellites were ve to six, and then covered good quality GPS (seven or more satellites) on a highway, as shown in Fig. 2(a). In addition, the second eld test

data [Fig. 2(b)] were collected on a different trajectory but with good satellite coverage. Normally, the GPS/IMU integrated navigation system has accurate trajectory, which is dominated by the GPS accuracy. To evaluate the performance of the IMU, the position drift of the IMU during GPS signal blockages is used as the key indicator. In this paper, 60s long GPS signal outages that cover different parts of the trajectory were simulated, as shown in Fig. 2. The IMU position errors during GPS signal outages are obtained by comparing the corresponding solution to the reference trajectory. The reference trajectory is acquired from the smoothed best estimate of a navigation-grade IMU and differential GPS (DGPS) data processed by Applanix Corporation POSPac software. The position errors showed the performance of ADI as a standalone system, and the comparison of the degradation due to misalignment was made on the basis of these position drift errors. The attitude errors were also considered as supplementary analysis tools to draw the conclusions. C. Analysis Strategy As the purpose of this paper is to evaluate the limits for the misalignment of the IMU to VBF without any knowledge of the presence of this error to the integration tool, the analysis will always be provided with respect to the drift errors of a well-aligned test IMU. These drift errors are given for GPS signal outage periods with respect to the reference trajectory. The following two different analysis criteria were used: 1) comparison of the maximum position drift errors during GPS signal outages between a misaligned (VBF to IMU frame) and a properly aligned test IMU; 2) comparison of the estimated attitude errors during GPS signal outages between a misaligned (VBF to IMU frame) and a properly aligned IMU. D. Different Sensor Combinations Two sensor combinations were investigated when IMU and VBF were misaligned. The original inertial data contain 16-bit resolution data recorded at a sampling frequency of 100 Hz using full IMU. In the search for optimal results in both economical and accuracy domains with the least resources and

3406

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 6, NOVEMBER 2008

TABLE I MODIFIED SENSOR SIGNALS AND THEIR PARAMETERS IN N AVIGATION A LGORITHM FOR 1G2A

Fig. 3. Different sensor combinations. (Darker shade shows gyroscopes and light shade represents accelerometers.)

partial conditions, a simplied sensor combination known as partial IMU was also used with one gyro and two horizontal accelerometers (1G2A), as shown in Fig. 3. The EKF equations remained the same, but before the start of processing, the omitted sensors are replaced by some assumed values, as given in Table I. The corresponding parameters in the navigation algorithm are also adjusted, including bias weighting, angle random walk (ARW), and velocity random walk (VRW) [17]. IV. R ESULTS This section is divided into two major parts. The rst part consists of the simulation results for the full IMU, and the second part comprises partial IMU results. Each of the parts is further divided into two subparts to show the results with and without nonholonomic constraints. A. Full IMU 1) Full IMU Without Nonholonomic Constraints: The results in Fig. 4 show the maximum position drift errors for simulated misalignments in three axes for the two data sets. The lines with circles represent the mean position drift when the roll misalignment is varied from 45 to +45 . The solid lines show the results for simulated pitch misalignments with respect to the no-misalignment case. The line with triangles demonstrates the position drift results with azimuth misalignment. In all cases, the average drift stayed within 2% of the perfectly aligned data set drift errors. The next set of results shows the ability of the EKF to estimate the induced attitude errors due to misalignment along each axis. First, misalignments from 45 to +45 were simulated in the roll angle, and the maximum error estimated by the EKF in roll, pitch, and azimuth attitude was investigated in Fig. 5(a) for the two data sets. The EKF showed an induced error of approximately 45 in roll angle (lines with circles) when the roll was misaligned to 45 . As this roll error is calculated with respect to a well-aligned inertial system roll angle, it, therefore, shows the ability of the full IMU to determine the simulated misalignment. The solid line and the line with triangles show the estimated errors in pitch and azimuth when the roll angle was misaligned. Fig. 5(b) shows the result for the two data sets when pitch misalignment was simulated. The result indicates that a misalignment in pitch will induce larger errors in roll and azimuth estimation as compared with the roll-misalignment case. The EKF was also able to estimate the induced misalignment in

Fig. 4. Drift errors for full IMU during GPS outages without nonholonomic constraint due to different misalignments.

pitch that was simulated in the data sets [solid lines in Fig. 5(b)]. Again, the misalignment errors are provided with respect to a well-aligned IMU in the same vehicle. The simulated azimuth misalignments are given in Fig. 5(c) with respect to well-aligned IMU navigation results. The navigation EKF estimations for the induced errors in roll, pitch, and azimuth show that roll and pitch are also exhibiting degradation when the azimuth was misaligned from 45 to +45 . The average induced errors for the two data sets due to simulated misalignments in roll, pitch, and azimuth are given in Table II. Although the specic values are different, which is mainly due to road conditions since both data sets were collected on different trajectories with different roll and pitch angles, the results can be summarized together. A misalignment in roll angle will minimally affect the estimation of the other two attitude angles as pitch and azimuth are important attitude angles for land vehicle navigation.

SYED et al.: CIVILIAN VEHICLE NAVIGATION: REQUIRED ALIGNMENT OF THE INERTIAL SENSORS

3407

TABLE II DEGRADATIONS OF FULL IMU (WITHOUT NONHOLONOMIC) UNDER DIFFERENT MISALIGNMENTS

data sets show a similar trend for a particular misalignment simulation, therefore, for simplicity, only typical results will be shown from this point forward. 2) Full IMU With Nonholonomic Constraints: The next set of simulations is conducted with nonholonomic constraints. Position drift errors and induced errors in the attitude angles due to misalignment in any of the axes are investigated. The application of these constraints for the perfectly aligned case signicantly reduces the position drift errors when compared with the results without constraints. Fig. 6(a) shows typical position drifts when misalignment along the roll (line with circles), pitch (solid line), or azimuth (line with triangles) axis was simulated. Fig. 6(b) depicts a typical example when misalignment was simulated in the azimuth axis, and it induced signicant errors in all attitude angles. Fig. 6(b) shows the estimated attitude errors by the EKF in roll, pitch, and azimuth when the azimuth axis was misaligned with respect to the VBF. This is in accordance with our previous analysis that the nonholonomic constraint strictly depends on the alignment of the IMU and, therefore, can only improve the navigation solution when the IMU is perfectly aligned with the VBF.
Fig. 5. Attitude errors induced by the simulated misalignment in the roll, pitch, and azimuth (continuous lines for data set 1 and dashed lines for data set 2). (a) Induced attitude errors due to roll misalignment. (b) Induced attitude errors due to pitch misalignment. (c) Induced attitude errors due to azimuth misalignment.

B. Partial IMU 1) 1G2A Without Nonholonomic Constraints: The rst set of results shows the position drift errors after individually simulating misalignment in the roll, pitch, and azimuth angles. Here, it is important to note that the position drift error for the perfectly aligned case of the partial IMU is larger than the position drift for any arbitrary case of the full IMU, mainly because of the missing sensors. An average 50% increase

Misalignments in pitch will affect both roll and azimuth estimations. Similarly, roll and pitch attitude angles will be affected for any misalignment in azimuth angle. Moreover, the average drift errors remain within 2% of the perfectly aligned system for any misalignment case. As it is evident that both

3408

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 6, NOVEMBER 2008

Fig. 6. Drift errors for full IMU during GPS outages with nonholonomic constraint under different misalignments. (a) Typical position drift errors. (b) Typical induced attitude errors due to misalignment in azimuth.

Fig. 7. Position drift errors for partial IMU (without nonholonomic constraint) under different misalignments. (a) Position drifts for data set 1. (b) Position drifts for data set 2. TABLE III DEGRADATIONS OF PARTIAL IMU (WITHOUT NONHOLONOMIC CONSTRAINT) UNDER DIFFERENT MISALIGNMENTS

in position drifts was observed for both data sets for the roll and pitch misalignment cases. The azimuth misalignment, on the contrary, only degraded the average position drift error by less than 10% (Fig. 7). The average percent increase in position drift errors is summarized in Table III. In addition, in the same table, the averages of the percent increase in attitude errors are provided. The table clearly indicates that only the simulated misalignment in azimuth can be closely estimated as the partial IMU only contains an azimuth-sensing gyro. Moreover, the effect of azimuth misalignment on roll and pitch angles is minimal as the EKF is able to properly detect the respective error in the azimuth. Fig. 8(a) shows the typical induced errors in all three attitude angles when roll was misaligned, whereas Fig. 8(b) shows typical resultant attitude angle errors when misalignments were simulated in azimuth. The average values given in Table III can easily be interpreted from these two gures. The nonsymmetry in Fig. 8(a) shows the inability of the partial sensors to properly estimate the roll misalignment errors. 2) 1G2A With Nonholonomic Constraints: As mentioned earlier, nonholonomic constraints require the correct knowledge of VBF to apply the physical constraints to the vehicle

motion. The application of these constraints improved the position drift errors from 254 to 76 m for data set 1 and from 421 to 127 m for data set 2 for perfectly aligned cases, which showed that it is important to use these constraints if a partial IMU is considered for economical reasons.

SYED et al.: CIVILIAN VEHICLE NAVIGATION: REQUIRED ALIGNMENT OF THE INERTIAL SENSORS

3409

Fig. 8. Typical induced attitude errors by misalignments in roll and azimuth. (a) Typical roll misalignment results. (b) Typical azimuth misalignment results.

Fig. 9. Position drift errors for partial IMU (with nonholonomic constraint) under different misalignments.

Fig. 10. Typical induced attitude errors by misalignments in roll, pitch, and azimuth. (a) Typical attitude errors due to roll misalignment. (b) Typical attitude errors due to pitch. (c) Typical attitude errors due to azimuth.

The rst set of results (Fig. 9) shows typical position drift errors when misalignments were simulated in the roll (line with circles), pitch (solid line), and azimuth (line with triangles) angles. Fig. 10(a)(c) shows the typical induced attitude errors due to misalignments in roll, pitch, and azimuth. The line with circles in Fig. 10(a) represents the range of estimated roll errors

due to misalignment simulation in roll. The error variation is small as the EKF cannot estimate the errors in the direction of the missing sensor. The solid line and line with triangles show the resulting attitude errors in pitch and azimuth due to the misalignment in roll. Fig. 10(b) and (c) shows the resulting attitude errors when misalignments were simulated in pitch and azimuth. The overall attitude errors due to misalignment in

3410

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 6, NOVEMBER 2008

each of the roll and pitch axes grossly degraded the errors in other attitude angles, mainly due to the lack of roll and pitch gyros. Consequently, the results are totally random for larger misalignment values in roll and pitch as the partial sensor suite cannot determine these misalignments. The results from Figs. 9 and 10 show that 15 misalignment in any of the attitude angles will increase the errors under 20% and can be considered toward the system design. V. D ISCUSSIONS The development of integrated GPS/IMU systems for civilian vehicle navigation is a challenging area, where cost, reliability, and ease of implementation are all important design aspects. This paper focuses on the requirement of alignment of the inertial unit with respect to the vehicle to produce reasonable navigation results for the most exible settings for the end user. Full and partial IMUs with and without nonholonomic constraints are investigated in the analysis. The analysis is performed during the GPS signal outages, where the inertial system works in standalone mode. In addition to the estimation of position drift errors, the errors induced in the other attitude angles due to the simulated misalignment are also studied in great detail. The comparisons for position drift errors and attitude angle errors are based on results from well-aligned IMU navigation results. The results and analysis given in this paper consisted of two data sets. Both data sets were collected on different roads with small but different roll and pitch angles. A. Errors Due to Misalignments This paper utilizes EKF since KF is considered as a preferred method for GPS/INS integration. The results obtained by using different sensor orientations [Fig. 1(b)] are compared with the results when the sensors were oriented exactly along the VBF [Fig. 1(a)]. This paper analyzes the errors due to erroneous information coming from the sensor due to misalignment and not the optimality of the EKF. Therefore, the same EKF is used to obtain the reference results with the nomisalignment case. This paper deals with the design of a system that can be used from one vehicle to the next without making special modications to the vehicle. In this kind of scenario, the orientation cannot always be guaranteed at every installation. The focus of this paper is to nd the system design specications for an approximate orientation for vehicle navigation. B. Full IMU Without Nonholonomic Constraints The misalignment in roll showed the least impact on pitch and azimuth estimation as roll is the least important in navigation. It clearly indicates that for full IMU, roll can be allowed to freely orient without any signicant effect on the position drift. From the design perspective of an easy operation civilian-grade integrated GPS/IMU navigation system, it sufces to say that roll can arbitrarily be oriented.

The pitch angle was subsequently misaligned, and the errors on position and other attitude angles were analyzed in detail. The maximum position drift error as compared with perfectly aligned IMU was less than 5%; however, the roll showed an average increase of 20% in the other attitude angle estimation. The maximum errors were in roll estimation for pitch misalignment greater than 25 . Pitch misalignment showed less impact on azimuth estimation errors. However, the average in azimuth estimation error was signicantly greater than the case when the roll angle was misaligned. The results indicate that if position drifts are the major requirement, pitch alignment is also not an important factor, and therefore, the consumer does not require the alignment of the pitch angle if full IMU is available. If correct attitude angles are required, the pitch angle should be constrained to 25 . The last set of results consisted of azimuth misalignment simulations. The maximum position drift increment stayed well under 10% of the no-misalignment case. However, the roll and pitch estimated errors showed serious degradations. A quick examination of Fig. 5(c) reveals that if correct attitude angles are required, the azimuth angle needs to be properly aligned. The navigation EKF was successfully able to determine the misalignment error in azimuth and, most importantly, was able to estimate the position drift of the vehicle during GPS signal outages for azimuth misalignment. Consequently, a full IMU can be used with no limit on the azimuth if roll and pitch accuracies are not important, which is the case for civilian navigation.

C. Full IMU With Nonholonomic Constraints Nonholonomic constraints required knowledge of the vehicle directions of motion, which in turn involved a well-aligned inertial system with respect to VBF. The application of nonholonomic constraints improved the position estimate for the perfect-alignment case. However, any misaligned attitude angle caused serious degradation in position drift errors and other attitude estimation. Therefore, for the best navigation results in civilian vehicle navigation, a well-aligned full IMU with nonholonomic constraints is the only choice. This kind of system is suited for a built-in navigation system in the vehicle but cannot be implemented in a portable device. D. Partial IMU Without Nonholonomic Constraints Any misalignment in roll and pitch angles signicantly degraded the results of the partial IMU. The azimuth misalignments, on the contrary, showed less than a 10% increase in position drift errors if nonholonomic constraint is not applied. The results presented here show that only the azimuth misalignment can be estimated and tolerated if a partial IMU is used. The partial IMU only has an azimuth-sensing gyro, and therefore, it is properly estimating the azimuth misalignments without inducing signicant errors in roll and pitch angle estimations. No error tolerance for misalignments in roll and pitch can be explained on the basis of only one gyro of the system that cannot estimate roll or pitch. This will also affect the estimation of pitch errors due to mountainous terrain.

SYED et al.: CIVILIAN VEHICLE NAVIGATION: REQUIRED ALIGNMENT OF THE INERTIAL SENSORS

3411

TABLE IV ALLOWABLE MISALIGNMENTS WITH DEGRADATION LESS THAN 20% (AS COMPARED WITH A WELL-ALIGNED SYSTEM)

R EFERENCES
[1] D. H. Titterton and J. L. Weston, Strapdown Inertial Navigation Technology. London, U.K.: Peregrinus, 1997. [2] C. Hide, T. Moore, and M. Smith, Adaptive Kalman ltering for low cost GPS/IMU, J. Navig., vol. 56, no. 1, pp. 143152, Jan. 2003. [3] X. Wang and G. Shen, A fast and accurate initial alignment method for strapdown inertial navigation system on stationary base, J. Control Theory Appl., vol. 3, no. 2, pp. 145149, May 2005. [4] Y. K. Peng and M. F. Golnaraghi, A vector-based gyro-free inertial navigation system by integrating existing accelerometer network in a passenger vehicle, in Proc. IEEE Position Location Navig. Symp., Monterey, CA, Apr. 2629, 2004, pp. 234242. [5] R. Wolf, G. W. Hein, B. Eissfeller, and E. Loehnert, An integrated low cost GPS/INS attitude determination and position location system, in Proc. ION GPS, Alexandria, Egypt, 1996, pp. 975981. [6] G. Dissanayake, S. Sukkarieh, E. Nebot, and H. Durrant-Whyte, The aiding of a low-cost strapdown inertial measurement unit using vehicle model constraints for land vehicle applications, IEEE Trans. Robot. Autom., vol. 17, no. 5, pp. 731747, Oct. 2001. [7] E.-H. Shin, Accuracy improvement of low cost IMU/GPS for land applications, Dept. Geomatics Eng., Univ. Calgary, Calgary, AB, Canada, UCGE Rep. 20156, Dec. 2001. [Online]. Available: http:// www.geomatics.ucalgary.ca/Papers/Thesis/NES/01.20156.EHShin.pdf [8] TomTom, Portable GPS Car Navigation Systems. visited on Oct. 3, 2007. [Online]. Available: www.tomtom.com [9] Sony, Personal Navigation, U.K. visited on Oct. 3, 2007. [Online]. Available: http://www.sony.co.uk/view/ShowProductCategory.action? site=odw_en_GB&category=ICN+Personal+navigation [10] Garmin, Well Take You There, visited on Oct. 3, 2007. [Online]. Available: http://www.garmin.com/garmin/cms/site/us [11] Navman In-Car & Personal Navigation, visited on Oct. 3, 2007. [Online]. Available: http://www.navman.com/Navman/templates/ BUHomepage____6313.aspx [12] M. S. Grewal, L. R. Weill, and A. P. Andrews, Global Positioning Systems, Inertial Navigation, and Integration. New York: Wiley, 2001. [13] X. Niu and N. El-Sheimy, Development of a low-cost MEMS IMU/GPS navigation system for land vehicles using auxiliary velocity updates in the VBF, in Proc. ION GNSS, Long Beach, CA, Sep. 1316, 2005. (CD10 p.). [14] S. Nassar, E-H. Shin, X. Niu, and N. El-Sheimy, Accurate IMU/GPS positioning with different inertial systems using various algorithms for bridging GPS outages, in Proc. ION GNSS 18th Int. Tech. Meeting Satellite Division, Long Beach, CA, Sep. 1316, 2005. [15] E. Nebot and H. Durrant-Whyte, Initial calibration and alignment of lowcost inertial navigation units for land vehicle applications, J. Robot. Syst., vol. 16, no. 2, pp. 8192, Feb. 1999. [16] R. G. Brown, Introduction to Random Signal Analysis and Kalman Filtering. New York: Wiley, 1983. [17] Z. Syed, P. Aggarwal, X. Niu, and N. El-Sheimy, Economical and robust inertial sensor conguration for a portable navigation system, in Proc. ION GNSS, Fort Worth, TX, Sep. 2425, 2007.

E. Partial IMU Without Nonholonomic Constraints The position drift errors of the partial IMU are signicantly larger, as compared with the full IMU; therefore, nonholonomic constraints should be used to get better navigation accuracies. In terms of misalignment, the results from Figs. 9 and 10 show that 15 misalignments will cause less than 20% degradation as compared with the perfectly aligned partial IMU and can be considered toward the system design.

VI. C ONCLUSION The results have shown that a full IMU with no nonholonomic constraint can deal with any kind of alignment error; therefore, if the ease of implementation from the users perspective is the most important criteria, a full IMU should be the system of choice. The partial IMUs show degradations when the roll and pitch angles were not aligned with the vehicle. The azimuth misalignment did not degrade the navigation accuracy for partial IMU. Nonholonomic constraints can improve navigation accuracy; however, these constraints cannot be applied if the IMU is not properly aligned with the vehicle. A full IMU is very sensitive to misalignment once nonholonomic is applied. On the contrary, a partial IMU with nonholonomic constraint can suffer misalignment of up to 15 if a degradation of 20% is allowed. Considering the overall tradeoff between performance, cost, and ease of use, the partial IMU with nonholonomic constraint is a promising solution. The results of this paper can be concluded using Table IV. It is important to reiterate that the issue of misalignment of the IMU with respect to the vehicle (Fig. 1) only arises if the navigation system is portable and the end user is responsible for properly mounting it.

ACKNOWLEDGMENT Dr. E.-H. Shin is acknowledged as the coauthor of the AINS toolbox used in this paper. The authors would like to thank H. Hou for helping during the eld test.

Zainab F. Syed received the B.Sc. degree in geomatics engineering and the M.Sc. degree in electrical engineering in 2001 and 2004, respectively, from the University of Calgary, Calgary, AB, Canada, where she is currently working toward the Ph.D. degree. She is also a Research Associate with Novatel Inc., Calgary. She has developed algorithms and different softwares for integrated inertial navigation, which is related to the development of a portable navigation system patent. This product is designed to overcome the shortcomings of existing navigation systems for personal use. She has provided technical support and guidance to clients including SiRF Inc., NemeriX, and Intermap Technologies for low-cost inertial sensor navigation. She has published 22 articles, including fully refereed journal papers, refereed conference papers, and conference proceedings.

3412

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 6, NOVEMBER 2008

Priyanka Aggarwal received the B.Sc. degree in electronics and communication engineering in 2000 from Kurukshetra University, Kurukshetra, India, and the M.Sc. degree in electrical and computer engineering in 2004 from the University of Calgary, Calgary, AB, Canada, where she is currently working toward the Ph.D. degree. She is also a Research Associate with AeroMechanical Systems, Calgary, where she has been developing a high-performance MEMS-based navigation system that fully integrates a GPS receiver chip and MEMS-based inertial gyros/accelerometer sensors to get accurate aircraft location and attitude. This work also involves decoding U-blox GPS receiver data. She has published 20 articles, including fully refereed journal papers, refereed conference papers, and conference proceedings in the eld of inertial navigation and MEMS-based microneedles. Her interests are primarily to integrate data from different sensors to accurately estimate the navigation states of a dynamic system using nonlinear lters, particularly particle lters.

Naser El-Sheimy received the B.Sc. (Hons.) and M.Sc. degrees from Ain Shams University, Cairo, Egypt, and the Ph.D. degree from the University of Calgary, Calgary, AB, Canada in 1996. He is currently a Professor and the Head of the Department of Geomatics Engineering, Schulich School of Engineering (SSE), University of Calgary, Calgary, AB, Canada. He holds the Canada Research Chair (CRC) in mobile multisensor systems. He has published a book and over 250 papers in academic journals and conference and workshop proceedings, in which he has received over 20 national and international paper awards. He has organized and participated in organizing many national and international conferences. His research interests include multisensor systems, mobile mapping systems, estimation techniques, real-time kinematic positioning, and digital photogrammetry and their applications in transportation, mapping, and Geospatial Information Systems (GISs).

Xiaoji Niu received the B.Eng. degree (with honors) in mechanical and electrical engineering and the Ph.D. degree in precision instruments from Tsinghua University, Beijing, China, in 1997 and 2002, respectively. From 2003 to 2007, he was a Post Doctoral Fellow with the Mobile Multi-Sensor Systems (MMSS) Research Group, Department of Geomatics Engineering, University of Calgary. He is currently a Senior Scientist with SiRF Technology, Inc., Shanghai, China. His research interests focus on lowcost GPS/inertial navigation technology for vehicles and pedestrians. Dr. Niu has been a professional member of the Institute of Navigation (ION) since 2003.

También podría gustarte