Pedestrian Inertial Navigation with Self-Contained Aiding
Description
Pedestrian Inertial Navigation with Self-Contained Aiding delivers a comprehensive and broad treatment of self-contained aiding techniques in pedestrian inertial navigation. The book combines an introduction to the general concept of navigation and major navigation and aiding techniques with more specific discussions of topics central to the field, as well as an exploration of the future of the future of the field: Ultimate Navigation Chip (uNavChip) technology.
The most commonly used implementation of pedestrian inertial navigation, strapdown inertial navigation, is discussed at length, as are the mechanization, implementation, error analysis, and adaptivity of zero-velocity update aided inertial navigation algorithms. The book demonstrates the implementation of ultrasonic sensors, ultra-wide band (UWB) sensors, and magnetic sensors. Ranging techniques are considered as well, including both foot-to-foot ranging and inter-agent ranging, and learning algorithms, navigation with signals of opportunity, and cooperative localization are discussed. Readers will also benefit from the inclusion of:
* A thorough introduction to the general concept of navigation as well as major navigation and aiding techniques
* An exploration of inertial navigation implementation, Inertial Measurement Units, and strapdown inertial navigation
* A discussion of error analysis in strapdown inertial navigation, as well as the motivation of aiding techniques for pedestrian inertial navigation
* A treatment of the zero-velocity update (ZUPT) aided inertial navigation algorithm, including its mechanization, implementation, error analysis, and adaptivity
Perfect for students and researchers in the field who seek a broad understanding of the subject, Pedestrian Inertial Navigation with Self-Contained Aiding will also earn a place in the libraries of industrial researchers and industrial marketing analysts who need a self-contained summary of the foundational elements of the field.
More details
Other editions
Additional editions
Persons
YUSHENG WANG, PhD, received the B.Eng. degree (Hons.) in engineering mechanics from Tsinghua University, Beijing, China, in 2014 and the Ph.D. degree in mechanical and aerospace engineering from the University of California, Irvine, CA, in 2020. His research interests include the development of silicon-based and fused quartz-based MEMS resonators and gyroscopes, and pedestrian inertial navigation development with sensor fusion. He is currently working at SiTime Corporation as an MEMS Development Engineer.
ANDREI M. SHKEL, PhD, has been on faculty at the University of California, Irvine since 2000, and served as a Program Manager in the Microsystems Technology Office of DARPA. His research interests are reflected in over 300 publications, 42 patents, and 3 books. Dr. Shkel has been on a number of editorial boards, including Editor of IEEE/ASME JMEMS, Journal of Gyroscopy and Navigation, and the founding chair of the IEEE Inertial Sensors. He was awarded the Office of the Secretary of Defense Medal for Exceptional Public Service in 2013, and the 2009 IEEE Sensors Council Technical Achievement Award. He is the President of the IEEE Sensors Council and the IEEE Fellow.
Content
Author Biographies xi
List of Figures xiii
List of Tables xix
1 Introduction 1
1.1 Navigation 1
1.2 Inertial Navigation 2
1.3 Pedestrian Inertial Navigation 5
1.3.1 Approaches 6
1.3.2 IMU Mounting Positions 7
1.3.3 Summary 8
1.4 Aiding Techniques for Inertial Navigation 9
1.4.1 Non-self-contained Aiding Techniques 9
1.4.1.1 Aiding Techniques Based on Natural Signals 9
1.4.1.2 Aiding Techniques Based on Artificial Signals 10
1.4.2 Self-contained Aiding Techniques 11
1.5 Outline of the Book 13
References 13
2 Inertial Sensors and Inertial Measurement Units 17
2.1 Accelerometers 17
2.1.1 Static Accelerometers 17
2.1.2 Resonant Accelerometers 19
2.2 Gyroscopes 21
2.2.1 Mechanical Gyroscopes 21
2.2.2 Optical Gyroscopes 22
2.2.2.1 Ring Laser Gyroscopes 22
2.2.2.2 Fiber Optic Gyroscopes 23
2.2.3 Nuclear Magnetic Resonance Gyroscopes 24
2.2.4 MEMS Vibratory Gyroscopes 24
2.2.4.1 Principle of Operation 25
2.2.4.2 Mode of Operation 25
2.2.4.3 Error Analysis 27
2.3 Inertial Measurement Units 28
2.3.1 Multi-sensor Assembly Approach 28
2.3.2 Single-Chip Approach 29
2.3.3 Device Folding Approach 30
2.3.4 Chip-Stacking Approach 31
2.4 Conclusions 32
References 32
3 Strapdown Inertial Navigation Mechanism 37
3.1 Reference Frame 37
3.2 Navigation Mechanism in the Inertial Frame 38
3.3 Navigation Mechanism in the Navigation Frame 40
3.4 Initialization 41
3.4.1 Tilt Sensing 42
3.4.2 Gyrocompassing 43
3.4.3 Magnetic Heading Estimation 44
3.5 Conclusions 45
References 45
4 Navigation Error Analysis in Strapdown Inertial Navigation 47
4.1 Error Source Analysis 47
4.1.1 Inertial Sensor Errors 48
4.1.2 Assembly Errors 51
4.1.3 Definition of IMU Grades 53
4.1.3.1 Consumer Grade 54
4.1.3.2 Industrial Grade 54
4.1.3.3 Tactical Grade 55
4.1.3.4 Navigation Grade 55
4.2 IMU Error Reduction 55
4.2.1 Six-Position Calibration 55
4.2.2 Multi-position Calibration 57
4.3 Error Accumulation Analysis 57
4.3.1 Error Propagation in Two-Dimensional Navigation 58
4.3.2 Error Propagation in Navigation Frame 61
4.4 Conclusions 62
References 63
5 Zero-Velocity Update Aided Pedestrian Inertial Navigation 65
5.1 Zero-Velocity Update Overview 65
5.2 Zero-Velocity Update Algorithm 68
5.2.1 Extended Kalman Filter 68
5.2.2 EKF in Pedestrian Inertial Navigation 70
5.2.3 Zero-Velocity Update Implementation 70
5.3 Parameter Selection 73
5.4 Conclusions 76
References 76
6 Navigation Error Analysis in the ZUPT-Aided Pedestrian Inertial Navigation 79
6.1 Human Gait Biomechanical Model 79
6.1.1 Foot Motion in Torso Frame 80
6.1.2 Foot Motion in Navigation Frame 80
6.1.3 Parameterization of Trajectory 81
6.2 Navigation Error Analysis 83
6.2.1 Starting Point 83
6.2.2 Covariance Increase During Swing Phase 84
6.2.3 Covariance Decrease During the Stance Phase 87
6.2.4 Covariance Level Estimation 88
6.2.5 Observations 92
6.3 Verification of Analysis 93
6.3.1 Numerical Verification 93
6.3.1.1 Effect of ARW 93
6.3.1.2 Effect of VRW 95
6.3.1.3 Effect of RRW 95
6.3.2 Experimental Verification 96
6.4 Limitations of the ZUPT Aiding Technique 99
6.5 Conclusions 100
References 101
7 Navigation Error Reduction in the ZUPT-Aided Pedestrian Inertial Navigation 103
7.1 IMU-Mounting Position Selection 104
7.1.1 Data Collection 105
7.1.2 Data Averaging 105
7.1.3 Data Processing Summary 107
7.1.4 Experimental Verification 109
7.2 Residual Velocity Calibration 110
7.3 Gyroscope G-Sensitivity Calibration 115
7.4 Navigation Error Compensation Results 117
7.5 Conclusions 119
References 119
8 Adaptive ZUPT-Aided Pedestrian Inertial Navigation 121
8.1 Floor Type Detection 121
8.1.1 Algorithm Overview 122
8.1.2 Algorithm Implementation 123
8.1.2.1 Data Partition 123
8.1.2.2 Principal Component Analysis 124
8.1.2.3 Artificial Neural Network 125
8.1.2.4 Multiple Model EKF 127
8.1.3 Navigation Result 129
8.1.4 Summary 130
8.2 Adaptive Stance Phase Detection 130
8.2.1 Zero-Velocity Detector 131
8.2.2 Adaptive Threshold Determination 131
8.2.3 Experimental Verification 135
8.2.4 Summary 136
8.3 Conclusions 138
References 139
9 Sensor Fusion Approaches 141
9.1 Magnetometry 141
9.2 Altimetry 142
9.3 Computer Vision 143
9.4 Multiple-IMU Approach 145
9.5 Ranging Techniques 146
9.5.1 Introduction to Ranging Techniques 147
9.5.1.1 Time of Arrival 147
9.5.1.2 Received Signal Strength 147
9.5.1.3 Angle of Arrival 148
9.5.2 Ultrasonic Ranging 149
9.5.2.1 Foot-to-Foot Ranging 150
9.5.2.2 Directional Ranging 150
9.5.3 Ultrawide Band Ranging 153
9.6 Conclusions 154
References 154
10 Perspective on Pedestrian Inertial Navigation Systems 159
10.1 Hardware Development 159
10.2 Software Development 161
10.3 Conclusions 161
References 162
Index 163
List of Figures
-
Figure 1.1 A schematic of gimbal system.
-
Figure 1.2 Comparison of (a) gimbal inertial navigation algorithm and (b) strapdown inertial navigation algorithm.
-
Figure 1.3 A comparison of (a) an IMU developed for the Apollo missions in 1960s.
-
Figure 2.1 The basic structure of an accelerometer.
-
Figure 2.2 Schematics of accelerometers based on SAW devices [11], vibrating beams [8], and BAW devices [9].
-
Figure 2.3 Typical performances and applications of different gyroscopes.
-
Figure 2.4 Schematics of a gyroscope and its different configurations [24]-[27].
-
Figure 2.5 Ideal response of a gyroscope operated in (a) open-loop mode, (b) force-to-rebalance mode, and (c) whole angle mode, respectively.
-
Figure 2.6 Schematics of two typical IMU assembly architectures: (a) cubic structure and (b) stacking structure.
-
Figure 2.7 Different mechanical structures of three-axis gyroscopes.
-
Figure 2.8 Examples of miniaturized IMU assembly architectures by MEMS fabrication: (a) folded structure and (b) stacking structure.
-
Figure 3.1 Block diagram of strapdown inertial navigation mechanism in the i-frame.
-
Figure 3.2 Block diagram of strapdown inertial navigation mechanism in the n-frame.
-
Figure 3.3 Relation between the gyroscope bias and yaw angle estimation error.
-
Figure 4.1 Common error types in inertial sensor readouts. (a) Noise, (b) bias, (c) scale factor error, (d) nonlinearity, (e) dead zone, (f) quantization.
-
Figure 4.2 A schematic of log-log plot of Allan deviation.
-
Figure 4.3 A schematic of the IMU assembly error.
-
Figure 4.4 Illustration of the two components of the IMU assembly error: non-orthogonality and misalignment.
-
Figure 4.5 Two-dimensional strapdown inertial navigation system in a fixed frame. Two accelerometers and one gyroscope is needed.
-
Figure 4.6 Propagation of navigation error with different grades of IMUs.
-
Figure 5.1 Relation between the volumes and the navigation error in five minutes of IMUs of different grades. The dashed box in the lower left corner indicates the desired performance for the pedestrian inertial navigation, showing the need for aiding techniques.
-
Figure 5.2 Comparison of the estimated velocity of the North and estimated trajectory for navigation with and without ZUPT aiding.
-
Figure 5.3 Diagram of the ZUPT-aided pedestrian inertial navigation algorithm.
-
Figure 5.4 Velocity propagation along three orthogonal directions during the 600 stance phases.
-
Figure 5.5 Distribution of the final velocity along three orthogonal directions during 600 stance phases. Standard deviation is extracted as the average velocity uncertainty during the stance phase.
-
Figure 6.1 (a) Interpolation of joint movement data and (b) simplified human leg model.
-
Figure 6.2 Human ambulatory gait analysis. The light gray dots are the stationary points in different phases of one gait cycle.
-
Figure 6.3 Velocity of the parameterized trajectory. A close match is demonstrated and discontinuities were eliminated.
-
Figure 6.4 Displacement of the parameterized trajectory. A close match is demonstrated for displacement along the direction (horizontal). Difference between the displacements along direction (vertical) is to guarantee the displacement continuity in between the gait cycles.
-
Figure 6.5 A typical propagation of errors in attitude estimations in ZUPT-aided pedestrian inertial navigation. The solid lines are the actual estimation errors, and the dashed lines are the uncertainty of estimation. Azimuth angle (heading) is the only important EKF state that is not observable from zero-velocity measurements.
-
Figure 6.6 Effects of ARW of the gyroscopes on the velocity and angle estimation errors in the ZUPT-aided inertial navigation algorithm.
-
Figure 6.7 Effects of VRW of the accelerometers on the velocity and angle estimation errors in the ZUPT-aided inertial navigation algorithm.
-
Figure 6.8 Effects of RRW of the gyroscopes on the velocity and angle estimation errors in the ZUPT-aided inertial navigation algorithm.
-
Figure 6.9 Relation between RRW of gyroscopes and the position estimation uncertainties.
-
Figure 6.10 Allan deviation plot of the IMU used in this study. The result is compared to the datasheet specs [13].
-
Figure 6.11 The navigation error results of 40 trajectories. The averaged time duration is about 110 seconds, including the initial calibration. Note that scales for the two axes are different to highlight the effect of error accumulation.
-
Figure 6.12 Ending points of 40 trajectories. All data points are in a rectangular area with the length of 2.2 m and width of 0.8 m.
-
Figure 6.13 Autocorrelations of the , , and components of the innovation sequence during ZUPT-aided pedestrian inertial navigation.
-
Figure 7.1 Possible IMU-mounting positions.
-
Figure 7.2 Noise characteristics of the IMUs used in the study.
-
Figure 7.3 Comparison of averaged IMU data and ZUPT states from IMUs mounted on the forefoot and behind the heel. Stance phase is identified when ZUPT state is equal to 1.
-
Figure 7.4 Navigation error of 34 tests of the same circular trajectory.
-
Figure 7.5 Comparison of estimated trajectories and innovations from IMU mounted at the forefoot (a) and the heel (b).
-
Figure 7.6 Experimental setup to record the motion of the foot during the stance phase.
-
Figure 7.7 Velocity of the foot along three directions during a gait cycle. The thick solid lines are the averaged velocities along three directions.
-
Figure 7.8 Zoomed-in view of the velocity of the foot during the stance phase. The light gray dashed lines correspond to zero-velocity state, and the dark gray dashed lines are the range of the velocity distribution.
-
Figure 7.9 Panel (a) shows the test statistics of the same 70 steps recorded previously. Thick solid line is an averaged value. Panel (b) shows the residual velocity of the foot along the trajectory during the stance phase. The inner, middle, and outer dashed lines correspond to threshold levels of , , and , respectively.
-
Figure 7.10 Relation between the underestimate of trajectory length and the ZUPT detection threshold. The thick solid line is the result of the previous analysis, and the thinner lines are experimental results from 10 different runs.
-
Figure 7.11 The solid line is an estimated trajectory, and the dashed line is an analytically generated trajectory with heading angle increasing at a rate of . Note that the scales for the and axes are different. The inset shows that the estimated heading angle increases at a rate of .
-
Figure 7.12 (a) Experimental setup to statically calibrate IMU; (b) experimental setup to measure the relation between gyroscope g-sensitivity and acceleration frequency [13].
-
Figure 7.13 Relation between the gyroscope g-sensitivity and the vibration frequency obtained from three independent measurements. The dashed line is the gyroscope g-sensitivity measured in static calibration. Inset is the FFT of the -axis accelerometer readout during a typical walking of two minutes.
-
Figure 7.14 Comparison of trajectories with and without systematic error compensation. Note that the scales for and axes are different.
-
Figure 7.15 Comparison of the end points with and without systematic error compensation. The dashed lines are the boundaries of the results.
-
Figure 8.1 Schematics of the algorithm discussed in this chapter. The numbers (1)-(4) indicate the four main steps in the algorithm.
-
Figure 8.2 An example of IMU data partition. Each partition (indicated by different brightness) starts at toe-off of the foot.
-
Figure 8.3 Distribution of eigenvalues of the centered data matrix after conducting the singular value decomposition.
-
Figure 8.4 Relation between the misclassification rate, PCA output dimension, and number of neurons in the hidden layer.
-
Figure 8.5 Confusion matrices of the floor type identification results with the PCA output dimension of 3 and 10, respectively. Classes are (1) walking on hard floor, (2) walking on grass, (3) walking on sand, (4) walking upstairs,...
