Professional Documents
Culture Documents
Navigatie Cu GPS
Navigatie Cu GPS
David M. Bevly Assistant Professor Department of Mechanical Engineering Auburn University, AL 36849-5341 Director of Auburn University's GPS and Vehicle Dynamics Lab (GAVLAB)
Presentation Overview
Overview of GPS (73 slides)
History of GPS Signal Structure Measurements and Accuracy IMU errors Introduction of Kalman Filtering JD and DGC Examples Navigation Errors
Lidar and Vision Navigation (30 Slides) GPS/INS for Estimation of Vehicle States and Parameters (30 Slides)
GPS and Vehicle Dynamics Lab 2
Motivation
Future stability control systems for passenger vehicles will use a precision navigation solution
Currently, a need for more vehicle information Also a need to improve accuracy of information Lane keeping systems can use position information
The DARPA Grand Challenge Military vehicles, especially Future Combat Systems (FCS)
Armored Robotic Vehicles (ARV) Robotic Armored Assault Systems (RAAS)
Control of Vehicles
need to know vehicle:
above measurements can be made using GPS can use the measurements (for example) to:
Coordinate Nomenclature
Vz , r , p Vx Vy p , q
V = velocity r = yaw rate p = roll rate q = pitch rate = yaw angle = roll angle = pitch angle = road grade
Coordinate Nomenclature
V = velocity r = yaw rate = heading (or yaw) = vehicle course = steer angle = body sideslip angle = tire sideslip angle
GPS Facts
There are more than 100 times as many civilian users than military users. 5 million recreational GPS devices were shipped in 2003, with a projected growth rate of 31% each year through 2009. Economics:
The cost of maintaining the GPS satellite system is $750 million each year, including replacing aging satellites. The direct economic impact of GPS is projected to exceed $50 billion by 2010
http://gps.losangeles.af.mil/jpo/gpsoverview.htm
GPS Signal
(satellite info)
19 cm
Roughly equivalent to viewing a 25-watt light bulb from a distance of 10,000 miles.
GPS and Vehicle Dynamics Lab 10
GPS Time
The continuous atomic timescale used on the satellites and control stations
Currently ahead of UTC time by 14 seconds Ex: UTC 10:34:25; GPS 10:34:39
11
3 Segments of GPS
Control Segment: 5 fixed location earth-based monitor stations
Stations located at: Colorado Springs, Ascension Island, Diego Garcia, Kwajalein, and Hawaii Responsible for maintain each of the satellites positions, clocks, etc. Track the GPS satellites and generate and upload the navigation data to each of the GPS satellites. Each satellite transmits at L1 (1575.4 MHz) and L2 (1227.6 MHz)
User Segment: all users, military and civilian, commercial and individual, who utilize the GPS signal
GPS and Vehicle Dynamics Lab 12
Coarse/Acquisition (C/A) and P(Y) Code Civilian Use P(Y) Code Military Use
L2 (1227.60 MHz)
13
L5
Civilian Signal broadcast at 1176.45 MHz Available 2015??
M-code
New military code
L1C (1st launch scheduled 2015) GALILEO (2010-2015 for full constellation) GLONASS (??) Australia, Japan, China, etc.
GPS and Vehicle Dynamics Lab 14
Other Countries
The Navigation message which includes the ephemeris data from the satellite is 30 secs. in duration and is transmitted in digital form at a rate of 50 bps. This data transmission modulates the GPS carrier wave using binary phase-shift keying (BPSK)
15
Allows satellites transmit on the same frequency at the same time without interfering with each other
GPS and Vehicle Dynamics Lab
16
Carrier Wave
L1, L2
Code Signal
C/A, P(Y)
Navigation Data
Satellite Information
17
Carrier Wave
L1 at 1575.42 MHz (154 x 10.23 MHz) L2 at 1227.60 MHz (120 x 10.23 MHz) Modulated with Code and Navigation Data using Binary Phase-Shift Keying (BPSK) C/A and P(Y) are transmitted orthogonally on L1
GPS and Vehicle Dynamics Lab 18
Code Signal
Code Division Multiple Access (CDMA) Course Acquisition - C/A
Gold Codes Code Period of 1 ms Anti-Spoofing Mode Code reset each week
GPS and Vehicle Dynamics Lab
19
Navigation Data
Navigation Data from the Data Bits
Clock Correction & Satellite Quality Ephemeris Ephemeris Almanac & Ionosphere & UTC Corrections Almanac
GPS and Vehicle Dynamics Lab 20
21
Satellite positions
deduced from Nav frame emphemeris & time
Pseudoranges
Measurement based on time delay from user to satellite
4
22
Pseudorange
Satellite Positions vs. Time Pseudoranges
Definition: iT
Range calculated by taking propagation time multiplied by speed of light Since clocks unsynchronized, clock errors are present -> pseudorange Measurement epoch occurs by shifting replicated code until correlation achieved (C/A code repeats every 1 ms)
= c(tu t si ) + cbu
iT + Di cbi + c(Ti + I i + vi + vi )
iT =
GPS Receiver
24
4. 5. 6.
From the navigation message, satellite position is calculated Using the transport time delay, Doppler frequency, and satellite position, the range to the satellite and velocity towards the satellite are calculated. By using 4 satellites or more, an extended Kalman filter or Least Squares algorithm combines the range and velocity to compute user position.
GPS and Vehicle Dynamics Lab 25
s = ( x s xu ) 2 + ( y s yu ) 2 + ( z s z u ) 2 + bs + s
Linearized measurement model is:
u , k 1 x s ,1 ) (x 1 1 , k 1 u , k 1 x s , 2 ) 2 (x . 2 , k 1 . . = . . . . x s ,m ) (x m u , k 1 m , k 1 u , k 1 y s ,1 ) (y u , k 1 y s , 2 ) (y u , k 1 z s ,1 ) (z u , k 1 z s , 2 ) (z
1 , k 1
1 , k 1
H k ( xu , x s ) s = * xk x k
2 , k 1
. . . u , k 1 y s , m ) (y
2 , k 1
. . . u , k 1 z s , m ) (z
m , k 1
m , k 1
1 1 x y 1 z 1 b 1 1
For 1,2m satellites -> m measurements Linearized about most current position estimate
GPS and Vehicle Dynamics Lab 26
Least Squares
xu y u x = zu ctu
x = H R H
T
H T R 1
1 = 2 3 4
a x1 a x2 H= ax3 a x 4
a y1 ay2 a y3 ay4
a z1 1 a z 2 1 a z 3 1 a z 4 1
u xj x axj = j r
i = r
u yj y a yj = j r
u zj z a zj = j r
(x
u )2 + ( y j y u )2 + (z j z u )2 x
GPS and Vehicle Dynamics Lab
j j =
27
= Hx
The position error can be estimated using LS as Calculating the covariance of x yields the 4x4 matrix D
x2 covariance terms 2 2 T y =2 D = UERE UERE H H 2 z 2 b covariance terms
x = H H
T
H T
28
29
DOP Values
Geometric (GDOP) Position (PDOP) Horizontal (HDOP) Vertical (VDOP) Time (TDOP) Note: TDOP is in m, not s
GPS and Vehicle Dynamics Lab
2 2 GDOP UERE = x + y + z2 + b2
30
Position Accuracy
Horizontal position accuracy is often assumed to have a bivariate Gaussian (normal) distribution x + 2 xy y
2
PDF ( x, y ) =
2 x
x,y
2 2 x y 1 x ,y
e
6 4 2 0 -2 -4 -6 y
2 2 1 x ,y
2 y
This results in probability ellipses Parameters come from solution calculation covariance
GPS and Vehicle Dynamics Lab
-5
0 x
31
Contains ~63-69% of the samples 2 DRMS contains ~95-98.5% of the samples Exact percentage within radius depends on the circularity of the ellipse (correlation coefficient) Closely matches Gaussian probability
GPS and Vehicle Dynamics Lab 32
Radius of a circle containing 50% of samples Originally used for military targeting accuracy
CEP 0.75 DRMS
Note:
80
Data CDF
Probability
60
40
20
0 0
r /
34
35
Difference in transmitted and actual satellite location (Slowly varying) SA contribution (now off) Based on stable atomic clocks
36
Free electrons cause delay of signal proportional to inverse of carrier frequency squared Without SA, the largest error component Requires model to correct Highly variable Smaller contribution to error Affects both L1 and L2 equally
GPS and Vehicle Dynamics Lab 37
Troposphere Errors
Reflected signals masking actual correlation peak Reduce by using cut-off angle, good antenna location, antenna and signal processing techniques Thermal noise Software accuracy
GPS and Vehicle Dynamics Lab 38
Receiver Errors
Parkinson and Spilker, Global Positioning System: Theory and Applications Vol. 1, AIAA, 1996 Ted Driver, Statistical Analysis of Military and Civilian Navigation Error Data Services, Proceedings of the 2006 ION-GNSS Conference
Error Source Ephemeris Data Satellite Clock Ionosphere Troposphere Multipath Receiver Measurement UERE, rms Filtered UERE, rms
GPS Errors
Receiver 1&2:RTD Receiver 3: Saphire Receiver 4: Starfire
41
GPS Errors
Common Mode Errors can easily be seen by two GPS receivers
42
43
44
Relative Positioning
Determination of the baseline vector between a known receiver location and arbitrary receiver location
If receivers are in close proximity (50km), they are subjected to very similar errors Differencing measurements from receivers removes errors, providing accurate baseline measurement
Carrier single differencing removes atmospheric errors and
satellite clock biases Carrier double differencing removes receiver clock bias Code double differencing removes atmospheric errors, receiver and satellite clock biases, and cycle slip effects (more noisy) Carrier triple differencing removes cycle slip effects
45
Absolute or relative differential positioning Local area, regional area, or wide Area Code based or carrier based
Kaplan Edition 2: Understanding GPS Principles and Applications
46
Use a Base Station (at known location) to correct common GPS errors
GPS and Vehicle Dynamics Lab 47
DGPS Accuracy
10 meter accuracy based on Federal Radionavigation Systems (FRS) report published jointly by the U.S. DOT and Department of Defense (DoD) Dependent on users distance from transmission source In 1993, the US DOT estimated error growth of 0.67 m per 100 km from the broadcast site Measurements of accuracy in Portugal suggest a degradation of just 0.22 m per 100 km
http://en.wikipedia.org/wiki/Differential_GPS
Recent results have shown troposphere errors can be significant in RTK systems over short baselines:
David Lawrence, et.al, Decorrelation of Troposphere Across Short Baselines, Proceedings of the 2006 IEEE/ION Positioning, Location, and Navigation Symposium (PLANS) GPS and Vehicle Dynamics Lab 48
NDGPS Coverage
49
Omnistar
Starfire
50
North (m)
10
15
20 20
10
10 East (m)
20
30
40
51
100
50
0 0
0.2
0.4
0.6 r (m)
0.8
1.2
52
53
N+
User
Reference Antenna
local reference station required solve for integer ambiguity track carrier phase phase at reference antenna is broadcast to user positioning v software calculates R 3-D accuracy* = 2 cm
More on RTK
L1 f=1.5 GHz
RTK Real Time Kinematic GPS RTK GPS calculates the relative position, R, between a rover and fixed base station to sub centimeter accuracy Integer ambiguity (IA), N, must be calculated
c = = 19 cm for L1 f
2 1 R= N + 360
Many published algorithms available Can take 20 minutes New techniques utilizing L1 and L2 (wide laning) are nearly instantaneous
L1 Signal 1 19 cm 2
55
Integer Ambiguity
The number of whole carrier cycles between a receiver and satellite
Carrier based DGPS technique utilizing the accuracy of a receivers phase measurement Estimates number of carrier cycles, N
Cycle slip is a sudden shift in the value of N when communication between satellite and receiver is compromised
30 minutes Dual frequency A wide lane or narrow lane approach can limit the possibilities of N, drastically reducing search time Triple frequency N can be nearly instantaneously solved using a third frequency, such as L5, GLONASS, or GALILEO
56
What is DRTK?
Dynamic RTK is the idea to use a moving, or dynamic, base station to calculate relative position between it and a rover L1 Signal
1 19 cm 2
57
Differences
RTK (Fixed base station)
IA algorithms are published Method well studied Cycle slip is minimized with good receiver Speed and distance ranges known
East v North = R = f ( N , ) Up
N+
User
Reference Antenna
59
Relative position is single vector from follower to moving base vehicle Global position is needed for path following
A vehicle might drop a temporary static base station The base vehicle could stop, forming a static base station Techniques utilizing relative measurement might be able to keep track of lead vehicle motion
GPS and Vehicle Dynamics Lab
60
Building DRTK
Building a DRTK system will allow modification of internal carrier tracking and IA algorithms Initial development with Novatel Superstar II receiver
61
global coverage global coverage not all are global, but almost full US coverage
local reference station ~0.3m Coast Guard differential corrections ~ 0.5m WAAS ~1-3m Nation Wide DGPS (NDGPS) ~ 1-3m OmniStar VBS (~1m) & Omnistar HP (~10cm) JohnDeere Starfire ~10cm
Accuracy
0.2-0.5 m/s with SA 3-5 cm/s without SA
=
vel
vel
vel
V
GPS
V = tan V GPS
1 up GPS
0.05 (rad) V
2 1.5 1 0.5 0 0
0 .1 (rad) V
10
15
20
25
30
35
40
65
Cohen C.E., Parkinson, B.W., McNally, B.D., Flight Tests of Attitude Determination Using GPS Compared Against an Inertial Navigation Unit, Navigation: Journal of the Institute of Navigation, Vol 41, No. 1, Spring 1994.
GPS Attitude
No Reference Station Required Accuracy Depends on Antenna Spacing (Not Velocity)
GPS Wave Front
66
67
Uses of GPS
?
GPS and Vehicle Dynamics Lab 68
69
Traffic Monitoring
Mean traffic position lies along the centerline of the lane
70
Pseudolites
Ground-based transmitter that emits GPS-like signals Pseudo-satellite -> pseudolite (PL) Many methods of implementation
C/A Codes Frequency Offset Pulsing Scheme Signal augmentation Data link enhancement
GPS and Vehicle Dynamics Lab 71
Objectives:
Pseudolite Types
Direct Ranging PL
Original (before GPS satellite) Satellite on the ground GPS scheme inverted (stationary receivers) Transmit data via GPS signal (max of 1000bps vs. 50bps GPS) Reflects message from GPS satellites
GPS and Vehicle Dynamics Lab
Mobile Pseudolite
Synchrolites
72
GPS References
System: Theory and Applications Volume 1&2, AIAA, 1996. Kaplan, Understanding GPS Principles and Applications, Artech House Publishers, 1996 Misra and Enge, GPS: Signals, Measurements, and Performance, Ganga-Jamuna Press, 2001
Parkinson and Spilker, Global Positioning
74
75
Motivation
Initial investigation of IMU models and IMU error sources Predict navigational accuracy during loss of GPS (function of IMU and dynamics of the trajectory). Understand the limits of GPS/INS performance (especially in advanced integration techniques such as ultra-tightly coupled GPS/INS).
GPS and Vehicle Dynamics Lab
76
LN200 IMU
[R ] = [R ][R ][R ]
Y
Y
Z
X
X
GPS and Vehicle Dynamics Lab
78
Moving bias, Turn On Bias, and Random Noise Moving bias always initialized to zero (since offset bias exits)
g r = r + c r + br + w gyro
E[br ] = 0 E b =
2 r
[ ]
2 gyrobias
& = b r
w gyro ~ N 0 ,
br + w gyrobias
fs
[ ]
2 gyro
w accel ~ N 0 ,
&x&
2 accel
fs
wgyrobias ~ N 0,
2 gyrobias
]
79
Other Accelerations
Accelerometers Measure specific force not true acceleration Must compensate for Gravity Field and Centripetal (and Coriolis) Accelerations
& + V y r + b x + Gc + waccel ax = & x
r = yaw rate = pitch = roll V = Velocity Gc = 9.81 m/s2
80
Inertial Sensor Error Modeling Using Allan Variance, by Hou and El-Sheimy, Proceedings of the 2003 ION-GPS Conference
Allan Variance
A time averaging technique used to determine error mechanisms. Viewed as the time domain equivalent of the power spectrum density
The PSD is the limiting mean square of a random variable Error Mechanism
Wide-Band Noise Exponentially Correlated Noise (First Order Markov Process) Rate Random Walk Linear Rate Ramp Quantization Noise Sinusoidal Input Flicker Noise GPS and Vehicle Dynamics Lab
AV Slope
- 1/2 1/2 1/2 1 -1 1 0
81
Autocorrelation Analysis
The expected value of the product of a random variable or signal realization with a time-shifted version of itself Used to determine time constant of stochastic process (bias drift) or a periodic nature in the signal
Simulated Sensor Data from MEMS Gyro
82
g r = r + c r + br + wgyro
0 insignificant
=
GPS and Vehicle Dynamics Lab
gyro
fs
83
Time (hours)
=
Time (hours)
GPS and Vehicle Dynamics Lab
gyro
fs
84
Take the Autocorrelation of the Filtered Data to determine the time constant of the Markov Process
Variance of the filtered data
2 E b&x2 = & accel bias
[ ]
bias
Time (hours)
accel
2 2 f s accel bias
&x&
&x&
Time (hours)
85
Used determined coefficients to generate a simulated sensor output Use an Allan variance to compare the simulated and experimental sensor outputs Shows that simulations can be used to generate realistic simulated data for navigation analysis and design
86
Attribute
Units
Spec
Attribute
Units
Specification
g/Hz sec g
rotation = gyro Ts t
Neglected bias but can be considered a best case scenario. Monte Carlo Simulation: Static Data Offset Bias Removed 1000 Iterations Sample at 100 Hz Monte Carlo simulation shows the effects of the walking bias
GPS and Vehicle Dynamics Lab 88
North
= waccel
1 3 Ts t 3
1 8
East
= waccel wgyro Ts t 2
Neglected bias but can be considered a best case scenario. Monte Carlo simulations validate the equations and show the effects of the walking bias
GPS and Vehicle Dynamics Lab 89
&+c +b + w g = gyro
Pitch Rate Gyro Model
&+c +b + w g = gyro
Yaw Rate Gyro Model
& + c + b + wgyro g =
90
& + sin ( Ax Ax )& & & + SFAz & & + SFN z & &2 + sin ( Ay + Ay )& a&z& = (1 + SFz )& z z z x y + c&z& + b&&x&& + waccel z GPS and Vehicle Dynamics Lab
93
myx Y
mzy Z X mxz
mzz
mzx
mxy
mxx
Nonorthogonality Errors
Large arrows represent the nominal axis (X,Y, and Z) Smaller arrows represent the nonorthogonality and scale factor errors
nyy
nyz Y
nyx
nzy
n y = arcsin xz X
n = arcsin zx Z
Z nxx
nzz
nzx
98
99
Scenario: Rocket Trajectory Duration is 165 seconds Monte Carlo Simulation 200 Iterations Constant Bias set to zero Initialization errors set to zero Shows that the additional terms in the advanced model only effect the mean impact point. Sigma bounds remain relatively equal
100
Contribution Level (meters) 500 450 400 350 300 250 200 150 100 50 0
The rocket trajectory was run with each error independently Errors that contribute the most are errors that see the accelerations and rotation rates
yr o Gy M i ro sa l M in Gy Gy r isa gm ro o M l ig en t Gy No isa nm ab o ro no r l ig e n t u t Gy No t ho nm a b o X ro no r go n en t u t No t ho a l a b Y no go it y ou t rt h na ab Z o li o R o go n t y a u t X ll ali b o Ac u P c e Y i t c Gy r t y a t Y a A le ro w h G o Sc b ou cc m Gy y ro al e t Z el et r o e e A S F Ac c c r om r M Sc a c a a c t e c e le e t e i sa l e le F o r F le ro r l a Ac ro m me Mi ign m a c to c t o c r r e te sa Ac e le r te r r M l ig ent Er c e o m No i sa nm a bo ro r le r e t no li e n ut o m e r rt gn t a X e te No h og men b ou r N no on t a t Y rt a b X o no h og li t y o ut Ac rth on a b Z a o Y c el e og o li t y u t X n Ac ro a a o Z c el m et li t y ut A er er ab Y c o X c e l m e S ca ou t Ac e r o t e r l e Z m S F c Y el e r e te ca l e a ct o r r A o Z c c m et Sc a Fa c A ele e r le to cc r A F r X e le om ss a c t Ac r o et e ym or m Y c el e et e r A s m et Ac ro r A sy ry m Z c el e m et ssy e t r A e cc rom r N mm y e le et on e t r o e r li n r y m N ea e t on ri er t l No i ne a y nl ri t in y e Al ar it lE y rro rs
101
The combination provides a high update rate, low noise, unbiased measurement solution
GPS and Vehicle Dynamics Lab 102
103
WB=1.0 m WB=0.007 m
GPS and Vehicle Dynamics Lab 104
Loosely Coupled Closely Coupled (Tightly Coupled w/out aiding) Tightly Coupled (w/aiding) Ultra-Tightly Coupled or Deeply Integrated
The methods differ in the type of information that is shared between individual units
Saurabh Godha, Strategies for GPS/INS Integration, http://www.geomatics.ucalgary.ca/~sgodha/Photos/GPS-INS.ppt
105
106
GPS Nav. Solution GPS Receiver Kalman Inertial Nav. Solution INS Filter Combined Nav. Solution
107
Cascaded architecture reduces the dimension of each state vector (less processing overhead) Easy to implement combine outputs from any commercial GPS receiver and IMU (dont need access to raw GPS measurements) Navigation solution relies on pure IMU measurements if GPS PVT is not available (number of satellites is less than 4) Correlated systems treated independently
Provides sub-optimal solution
GPS and Vehicle Dynamics Lab 108
Disadvantages
109
INS
110
Allows for continuous (degraded) positioning even when the number of satellites of GPS drops below 4 Allows for monitoring of individual GPS measurements from each satellite IMU aiding of tracking loops can improve GPS tracking in high dynamic environments More difficult to implement Larger size of state vector in centralized KF requires more computation time Provides no long-term noise immunity to GPS reception
GPS and Vehicle Dynamics Lab 111
Disadvantages
Inertial measurements are combined with the GPS signal measurements at the tracking level Raw GPS and IMU measurements are combined in a centralized navigation filter Filter operates on the receiver tracking loop I and Q signals and the IMU measurements in order to estimate navigation information
Requires access to receiver tracking loops or raw IF Also known as Ultra Tight Coupling
GPS and Vehicle Dynamics Lab 112
NCO
Frequency Estimate
INS
113
Can provide improved accuracy Increased jamming resistance Allows faster signal acquisition and reacquisition Provides improved tracking of GPS signal in the presence of high noise and/or high dynamics Currently not robust (no method for integrity monitoring of individual satellites since raw data is fused a in single navigation filter) Extremely cumbersome Sensitive to IMU noise and bias as well as method of implementation
GPS and Vehicle Dynamics Lab 114
Disadvantages
GPS/INS Integration
GPS/INS integration is predominately performed using Kalman Filters
Optimal fusion of data given sensor statistics Extended Kalman Filters (EKFs) required when problem becomes non-linear
Coordinate Transformations Estimating Certain IMU Scale Factors
Linear about small orientation angles Only estimated additive IMU errors (bias drift) Unscented Kalman Filters Particle Filtering Etc.
GPS and Vehicle Dynamics Lab 115
[ ] E [ww ] = Q
T T
w R d 1 Qc R d d v R m1 Rc R mm wk R n1 Qd R nn vk R m1 Rd R mm
d: # of disturbances n: # of states m: # of measurements
Discrete/Sampled
xk +1 = Ad xk + Bd u + wk yk +1 = Cxk + vk
E vk v k
k
[ ]= R E [w w ] = Q
d k
116
Process Disturbance
Ew
[ ]
2 vel
Rwpsd = Qc = 0
0 b2a
2 Rwpsd = Ts accel
tk
Qc T T w Ts Bw Qc Bw Ts
2 0 Ts accel 1 Tb 1
Qd =
t k 1
BwQc B (e ) d
T w
A T
0 1 b2a 0 0
0 1 Tb
Byrsons Trick:
c11 c12 c= =e 0 c22
T Ad = c22
T A BwQc Bw Ts AT 0
2 Ts accel 1 0 = Ts 1 0 1
1 0 2 ba 0 1 2 Tb
T Qd = c22 c12
Qbias =
b2
Tb2
117
]
118
E wwT = Q( )
T k)
] E [vv ] = R(t
T
& f x =A = x x h =C x
k 1
k T T d A( ) P( ) + P( ) A( ) + BwQ( ) Bw t k 1
1
)( x x ) = P(t k ) E (x x
Use numerical integration (Euler, Runge-Kutta, etc.) to propagate states and error covariance matrices
GPS and Vehicle Dynamics Lab 119
Time update integrates IMU measurements between GPS updates KF filters integrated IMU and GPS measurement at every GPS measurement
Based on predicted error from propagated IMU between GPS
Measurement Update
1 T T Lk = Pk C (CP C + R ) k v X k = X k + Lk ( y meas CX k n ) Pk = ( I Lk C ) Pk
k +1 = x k + u k x Pk +1 = Pk T + Qd
120
If GPS is available:
0 w 1 vel Tb
C = [1 0]
# Lk = #
[ ]
Rwpsd = Qc = 0
0 b2a
2 Rwpsd = Ts accel
VGPS
& x = C + vGPS bx
2 2 E vGPS = Rd = GPS
[ ]
P P = 11 P21
GPS and Vehicle Dynamics Lab
[ ] ) [ ] E ( b
( )
b & Ex x
x
121
= vel =
Ew
)
0 b2g
0 w 1 hd Tb
2 = Rv = E v
[ ]
2 hd
2 Ts gyro = Qc = 0
GPS = GPS
= [1 0] vel + v g bias
[ ]
2 GPS
vel Turn off KF during turning/periods of changing sideslip: = GPS (compares GPS velocity, course, with integrated gyro)
122
Longitudinal Dynamics
System Model
0 G c 0 = 0 0 0 0 0
Gc 0 0 0
0 V x 1 0 g + 0 1 ( p + bx ) 0 1 b Tb q 0
0 1 a x 0 0 + 1 g q 0 0 0
0 0 w 1 0 long 0 1Tb 0 0
GPS
0 0 0 b2g
123
2 E vGPS
[ ]
2 GPS = Rv = 0
2 GPS ( up ) V 0
2 E wlat
[ ]
2 Ts accel = Qc = 0 0
2 Ts gyro 0
Lateral Dynamics
System Model
Cannot distinguish pitch and longitudinal accelerometer bias Must account for centripetal acceleration
0 G C 0 = 0 0 0
1 0 V y 1 0 ( ) ( ) a g b V r r x y 1 ( + b y ) + + 0 1 0 g p 0 1 bp 0 0 Tb
0
1 Tb
wlat 1 Tb 0 0
= GPS vel
2 E wlat
[ ]
2 2 Ts accel + V 2 gyro = Qc = 0 0
0
2 Ts gyro
0 0 b2g
2 2 E vGPS = Rv = GPS
[ ]
124
Complimentary Filters
Can use complimentary filters to separate low frequency accelerometer bias from higher frequency vehicle dynamics KF Corrects IMU Errors
= Tb s ( + b ) y Tb s + 1 = b y 1 ( + b y ) Tb s + 1
Ts = b ( + bx ) Tb s + 1 = b x 1 ( + bx ) Tb s + 1
125
126
127
& (
& =0 ) = 0 or V y
128
(deg)
1 0
1 0 3 2
10
20
30
40
50
60
70
(deg)
1 0
1 0
10
20
30 Time (s)
40
50
60
70
Model Estimated
0.4
V (m/s)
0.2 0
V (m/s)
10 20 30
0.5 1
0.2 10 20 30 40 0 0.2 10 20 30 40
1 0
1.5 0 2
b and b (deg/s)
b and b (deg/s)
10 Time (s) 20 30
+ bx (deg)
+ bx (deg)
0.2
5 0 10 20 Time (s) 30 40
0.4 0
10
20 Time (s)
30
40
1 0
133
0.2
(deg)
0.2 0
P0.5
GPS
10
20
30
40
50
0 0
10
20
30
40
50
=0.05 m/s2
0.2
(deg)
0.05
est
0.1
1000
10
20 30 Time (s)
40
50
134
User with High Dynamics requires the Fusion of GPS & IMU
GPS gives:
User position in global coordinates (loosely coupled) Pseudorange measurements, satellite positions, & time all in global coordinates (tightly coupled) Acceleration Rotation Rate
135
x 1 y = 0 0 z
0 1 0
0 0 1
0 0 0
0 0 0
x y 0 z 0 + . 0 . .
s =
H k ( xu , x s ) * xk x k
i = 1, 2 ,... m
For m satellites in view (w/ valid pseudorange observations) State vector x includes states for the additional IMU dynamics
GPS and Vehicle Dynamics Lab 136
Summary of Methods
Loosely Coupled
GPS
, rsats
Least-Squares
IMU
a,
User States
Tightly Coupled
GPS
IMU
User States a,
137
138
1 , k 1
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
2 , k 1
. . . u , k 1 , y s , m ) (y
m , k 1
1 N E 1 V 1 1 b a 1 b 1 b gps 1
1 2 . + . . . m
139
a 2 0 Qc = 0 0
2
0 0
0 0
ba 2
0
0 0 0 b 2
140
good satellite visibility on farms relieve drivers from tedious & monotonous labor provide farm operation during poor visibility open doors for new agricultural techniques
GPS and Vehicle Dynamics Lab 141
Automated Steering
Cm-level control for row crops Reduced overlap for tillage Cooperating Vehicles
142
143
144
~ 3 meter lever arm 0.4 roll accuracy required to utilize 2 cm DGPS position accuracy
145
a C f s +
a C f c1C f
Kinematic Model
R= Vx L
146
Magnitude
0 10 20 0 10 0
10
Phase (deg)
50 100 150 10
0
10
Vx= 4 m/s
R (t ) =
147
Line Tracking
0.9 Later Error (m) 0.6 0.3 0.0 -0.3 0 50 100 Time (sec) 150 200 Mean = 5 mm 1=3 cm 1 Foot
148
Advanced Trajectories
149
Vx=8 m/s
Lateral Error (m)
0.2 0 0.2 0 0.4 Mean=3.5 cm 1=4.0 cm
Mean=2 cm
1=2 cm
2 1=0.08
10
10
15
20
Control Input
Control Input
1=0.10
4 Time (sec)
10 Time (sec)
15
20
150
N VX
& && b
& b
gb
rb
E = tractor east position N = tractor north position V x = forward velocity = heading & = yaw rate && = yaw acceleration b = slip angle (heading bias or " crab angle" ) = steer angle & = steering slew rate b = steer angle bias g b = gyro bias rb = radar bias
GPS and Vehicle Dynamics Lab 151
Bias Estimate
Tractor Model
LQR Control
Separate Estimators
Dead Reckoning Equations
& = ( radar r ) sin( ) e b & = ( radar r cos( ) n b) & = gyro g b
Inputs: Radar, Gyro
X 1 = e n rb
Tractor Equations
&&& = 2 && &+K n K d v & & & = + vu I I v v 2 n 2 R n
gb b
&& X2 =
& VX
& b
153
Able to accurately estimate both the gyro and steer angle bias independently
154
3 Tests VX=2 m/s Line Tracking Results Errors < 0.3m for 40 sec Errors <
Error Analysis:
& E = V X E = V X & TS t y
y (t ) = V X & TS * t
2 3
3 2
155
0.3 m
Test VX=2 m/s Line Tracking Results 8 sec < 9 cm error 42 sec < 30 cm error => % Error Lateral Errors > Longitudinal Errors
156
0.5 Lateral Error (m) 0.0 -0.5 -1.0 -1.5 -2.0 0 20 Gyro Heading
GPS Heading
GPS vs. GYRO < 0.3 m Crab Angle < 1 @ t=20 sec
40
60
80
157
1 0.23 m 0.87
Implement Control
4 Added States
X = e
n b ]
GPS Measurement
( s)
a 1+ b KI = = R( s ) s + VbX s + I
Implement Control
Implement: 7.9 m (26 foot) Wide Chisel Plow Implement Position: Carrier Phase DGPS 3-D Position (1 = 2 cm)
GPS Antenna (L=6.5 m)
160
70 65 60 55
North (m)
Implement Tractor
North (m)
2 1 0 1 2 10
10
20 30 East (m)
40
30
0 East (m)
10
15
20
161
Analytical Models
Can a model capture the behaviors actually seen from real data? Models useful to steering control describe the turning rate of the tractor from a given steering angle
c Fyi Fyr b
Fy = C
a Fyf Vx
i
Vit
r
Vrt Vy
V
f
Vft
Fxf
162
Empirical Models
Used to determine how a tractor really behaves with changes in implement Determine appropriate analytical model (and parameter variations) using experimental data
163
164
SciAutonics
Terramax
166
The incorrect model results in instability A better model is needed before the other measurements can be utilized
167
Tactical grade inertial measurement unit provided high update accelerations and angular rates
Rockwell Collins GIC-100 (<1/hr drift rate) TCM2 (16 Hz output) Microstrain 3DM-GX1 (50 Hz output)
Navigation
Navigation solution blends measurements from various sensors -DGPS -Steer angle -Multi-antenna GPS -IMU -Wheel speed -Lidar -Doppler radar State estimates provided to controller: -Velocity -Course -Position -Pitch -Roll -Measurement biases Other sensors can be easily added -Range radar -Camera -Ultrasonic
169
2 2 , t
Sensors can be modeled with a turn on bias, a Markov bias, and white noise
~ [0,1]
GPS offers precise information, but only at low update rates of 5 Hz IMU provides 50 Hz measurements, but suffers from bias drift Magnetometers provide roll, pitch and yaw measurements, but contained quickly drifting bias TCM2 output at 16 Hz Speedometer had a calibration error and was susceptible to wheel slip Also output at a variable rate A Kalman filter was used to blend the various sensor measurements and provide reliable information based on the strengths of each sensor while compensating for their inadequacies
GPS and Vehicle Dynamics Lab
171
Inputs were from IMU Biases were not modeled as a function of time, but the noise driving the drift was modeled in the process covariance matrix
Q d = diag ax r & = 12 b + w2 b
2
br 2 N 2 E 2 2 b 2 2 ... bM 1 2 bM 2 2 bM 2 2 bM 2 2
b g
bM 1 2 bM 1 2
y = [VGPS
VWS GPS
N GPS
EGPS
Velocity state accounts for vehicle pitch and longitudinal road grade Vehicle pitch estimate contains longitudinal accelerometer bias Roll estimate is vehicle roll and lateral road grade, and contains lateral accelerometer bias
GPS and Vehicle Dynamics Lab 172
173
Settle time when GPS is acquired ~ 3 seconds Algorithmic logic in the loop leads to better navigation solutions
Some commercial systems take >30 minutes to initialize Magnetometers aided initialization, but were statistically weighted out of the filter when the vehicle was moving
Yaw (deg)
Fast bias drift makes measurements unreliable Passing metallic objects also degrade quality of measurement
20
70
80
90
110
120
174
175
Bias estimates remain constant when no measurements exist to update them Integration of constant offset linearly degrades heading estimate Tactical grade IMU with low bias drift allowed for relatively long dead reckoning periods when the bias was correctly estimated before the outage
Calibration error in wheel speed sensor created an offset in the velocity estimate
Wheel slip would also introduce estimate error If the wheel speed bias was estimated, the calibration error and wheel slip would corrupt it as well
GPS and Vehicle Dynamics Lab 177
~1m error after 25 seconds Error Caused by neglected sideslip generated during cornering
-10 -15 -20 North (m)
Error (m) 1 0.8
0.6
0.4
0.2
70 80 90 100 110 -0.2 East (m) 380 390 GPS and Vehicle Dynamics Lab GPS EKF
400
430
440
450
178
A single GPS antenna measures the direction of travel An integrated yaw rate gyro yields the direction the vehicle is pointing
150
N
Heading (deg)
100
GPS KF
VF
F
Vx r VR
R
20 Error (deg)
E
10 0
Vy
-10 GPS course is denoted by -20 Heaind is denoted by and Vehicle Dynamics 85 GPS Lab
90
95
100 Time
105
110
179
115
Navigation Errors
Typical kinematic navigation models neglect the lateral & = 0 and = 0 r dt vehicle dynamics and assume
Estimates velocity, course, position, and sensor biases (sometimes roll) Examples shown use IMU as input, and GPS as measurement
a x b ax 0 r b r & = x 0 cos( ) V ) V sin(
Model 1:
V b ax = x b r N E
V y= N E
Model 2:
V b ax b r = x +b ay b N E
V y = M N E
M =
1 ) a y V (r b r g
180
Navigation Errors
When sideslip is present, typical navigation models break down Expanding the model to include the lateral dynamics reduces estimation error in the presence of sideslip
Requires measurement of sideslip (2 antenna GPS receiver) Also provides lateral velocity estimate, which can be useful to stability control systems
+V ) (r b ax b ax y r 0 V ) g sin( a b ) (r b ay x r y 0 r b r & = x 0 & b IMU 0 2 2 1/ 2 /V )) + atan (V y x (V x + V y ) cos( 2 +V 2 )1 / 2 sin( (V + atan (V y / V x )) y x
V x bax V y bay = x b r b N E
Model 3:
V x V y y= V 1 y = tan V N x E
181
Navigation Errors
Model Comparison
Models 1 & 2 do not track vehicle course or vehicle heading Model 3 tracks vehicle heading well Can effect dead reckoning performance
1&2 3 True Course True Heading
2 1.5 1 0.5 0 -0.5 1&2 3 True br
10
10
15
20
25
30 Time (s)
35
40
45
50
182
Navigation Errors
Model Comparison
Dead reckoning performance of Model 3 is also better GPS artificially removed during the maneuver generating sideslip
Worst case scenario because bias estimates contain large errors Position model incorrect because it is integrating heading, not course
70 60 50 Position Error (m) 40 30 20 10 0 GPS and Vehicle Dynamics 0 Lab 10 20 Time (s) 30 40 50
Model 1&2:
= V N cos( ) = V E sin( )
Model 3:
= V N cos( + ) = V E sin( + )
1&2 - Outage During Slip 1&2 - Outage Before Slip 3 - Outage During Slip 3 - Outage Before Slip
183
Navigation Errors
Model Comparison
Model 3 tracks heading relatively well, but models 1&2 break down when the slip first occurs
-75 Yaw Estimate (deg) 1&2 3 True Course True Heading
-80
-85
-90
-95
GPS 197.5
199.5
184
Navigation Errors
Model Comparison
To improve navigation accuracy, the model must account for lateral vehicle dynamics
95 Yaw Estimate (deg) 90 85 80 75 1&2 3 True Course True Heading Position Error (m) 120 100 80 60 40 20 0 1&2 3
116
117
118
121
122
40
60 80 Time (s)
100
120
185
Velocity, local heading Local lateral error (for use in some controllers) Lateral vehicle movement (also makes vehicle sideslip and/or lateral velocity observable)
Navigate relative to the defined corridor Preliminary experiment was run indoors in a hallway
GPS and Vehicle Dynamics Lab 186
ylat2 2 1
ylat1
ydr, 0
Using the 0 measurement, lateral position (in the vehicle y = [y ] frame) to the wall is directly measured u = [a x ]
d1
One did not use a lateral accelerometer input The other used an ideal accelerometer input Both used the same measurements
Lidar/IMU Equations
No Lateral Accelerometer (Model 1)
& V x a x b ax & b 0 ax & & = r b 1 = x r & 0 b r sin( ) V & y lat x
ax 2 0 Q1 = 0 0 0 0 0
2
a u1 = x r
a x u 2 = r a y
0
2
bax
0 0 0
r
0 0
0 0
br
0
0 0 0 0 ylat 2
ax 2 0 2 0 bax 0 0 Q2 = 0 0 0 0 0 0
0 0
0
2
0 0 0 0
r
0 0 0
0 0
br 2
0 0
ylat 2
0
0 0 0 0 0 2 ay
189
With the measurements from the laser scanner, usable estimates are obtained
2 Lidar r KF 1.5
Lidar ax KF
Velocity (m/s)
0.5
Removes inertial sensor errors to provide unbiased solution Velocity and heading cleanly estimated
10 Time (s)
15
10 Time (s)
15
20
-0.05
-0.1
-0.15 0
10 Time (s)
15
190
Accounts for unmodeled dynamics and disturbances, such as lateral velocity Level of filtering on estimate decreased with higher values for ylat
0.1 0 -0.1 Lateral Error, ylat (m) -0.2 -0.3 -0.4 -0.5
0.1 0 -0.1 Lateral Error, ylat (m) -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 0 Lidar KF 5 10 Time (s) 15
-0.6
Lidar KF
10 Time (s)
15
20
191
20
192
Research Motivation
Vehicle lane departure major cause of highway fatalities
42,000 roadway fatalities in 2004, 50% resulting from vehicle lane departure ADAS Advanced Driver Assistance Systems LDW- Lane Departure Warning
Use camera to detect a lane marker Send warning to driver if lane is being approach Helps to prevent un-intended lane departure
ITS Research
1 0 1 G x = E 2 0 2 1 0 1
1 2 1 Gy = E 0 0 0 1 2 1
195
Edge Detection
Compute a gradient magnitude
G = Gx + G y
2 2
Threshold the magnitude to define a detected edge 1, G > t g E= o . w. 0, Canny edge detection extension of Sobel operator, most common method
GPS and Vehicle Dynamics Lab 196
Corner Detection
Takes edge detection one step further General method
1. 2. 3.
Gradient filtering (2D-convlution) Extract the edge image Find corners along detected edges
197
Demonstration
Stop motion sequence Top frame corners marked with white dots Bottom frame edge detection with Sobel operator
198
Optical Flow
Optical flow computation gives a velocity estimation of objects within an image frame Horn and Schunck - Least squares minimization with constraints on smoothness and brightness Three dimensional movement projected onto 2D image plane
GPS and Vehicle Dynamics Lab 199
200
Performance Limitations
Current LTCSs suffer from:
Environmental constraints Occluded markers: rain, mud, snow Lighting: poor lighting (night), glare from sun Highway conditions Poor/non-existent lane markings Markers occluded by another vehicle
Our Research
Improved LTCS Fuse GPS/INS/MAP with camera system Assist and/or compensate when lane detection fails
GPS and Vehicle Dynamics Lab
201
Simulation Experiment
Vehicle model generate, East, North and heading Create GPS, IMU and camera measurements from data Simulate vehicle driving in a mapped road lane
202
Tests conducted at National Center for Asphalt Technology Test Track (w/ Road Database
GPS and Vehicle Dynamics Lab 204
Differential GPS
Non-Differential GPS
205
10
10
15 800
206
1 1.5 2 2.5 3 3.5 800 700 600 500 400 300 200 Longitudinal distance from reference (m) 100 0 100
2 Error 1.5 1
Error
GPS and Vehicle Dynamics Lab 500 400 300 200 100 Longitudinal distance from reference (m)
207
Inputs
Lateral position and velocity Heading angle Biases (accelerometer, gyro, GPS) Accelerometer Gyro Lateral offset (camera) Lateral offset (GPS + map) Heading (GPS)
Measurements
GPS = +GPS
Measurements:
Inputs:
a accel = a y + c a + b a + w a rgyro = r + b r + c r + w r
208
Estimation Model
& 0 V y 1 & y & 0 & = 0 ba b & 0 r 0 & bGPS 0 0 0 0 0 0 0 0 0 0 0 0 0 V 1 y 0 0 y 0 0 1 0 0 0 0 0 b + 0 a 0 0 0 b 0 1 r 0 0 bGPS 0 Vy y 0 0 LTC 0 1 + GPS 1 b a 0 0 b GPS 2 r b GPS 0 1 0 0 0 1 0 0 0
& = Ax + B u u + B w w x
y = Cx +
ca 1 0 0 a y cr 0 r + 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 wa 0 w r 0 wb 0 a wb 0 r wGPS 1
0 y= 0 0
1 1 0
0 0 1
0 0 0
2 a 0 0 0 0 y 2 0 0 0 0 r 2 Qc = 0 0 ba 0 0 2 0 0 0 0 br 2 0 0 0 0 bGPS
2 LTC Rd = 0 0
0
2 GPS
y
0 0 2 GPS
209
Experimental Data
Data collected on south side of NCAT
Data logged at 70 Hz
210
GPS/LDW/IMU Measurements
Course/Yaw
GPS and Vehicle Dynamics Lab
Left Lane
Right Lane
211
Results
Estimation of IMU Biases and GPS/Lane Database Offset
212
Camera failure
Camera failure (30 s outage rely on corrected GPS/Map database) Incorrect lane detection
213
Orientation Accuracy
Determined using a covariance analysis based on sensor noise statistics
IMU Only
0.15
IMU/Vision
0.08 0.06
0.1
0.04
0.05
0.02
(deg)
(deg)
0 0.02 0.04
0.05
0.1
0.06
0.15
0.2 0
10
20
30
40
50 60 Time (s)
70
30
40
50 60 Time (s)
70
80
90
100
214
Velocity Accuracy
Determined using a covariance analysis based on sensor noise statistics
IMU Only
2.5 2
0.3 0.4
IMU/Vision
1.5
0.2
1 0.5
0.1
(m/s)
V (m/s)
0 0.5 1
0 0.1 0.2
0.4
30
40
50 60 Time (s)
70
80
90
100
215
Position Accuracy
Determined using a covariance analysis based on sensor noise statistics Assumes data is available (data base) to produce a vision based position solution
IMU Only
80 60
IMU/Vision
1.5 1
40
0.5
20
(m)
(m)
20
0.5
40
1
60 80 0
10
20
30
40
50 60 Time (s)
70
1.5
30
40
50 60 Time (s)
70
80
90
100
216
State estimation can then be used to estimate critical vehicle parameters for steering control systems
GPS and Vehicle Dynamics Lab 217
FyOF
x FyIF
V OF OF FxOF
V VIF IF FxIF
V = velocity r = yaw rate = vehicle heading = steer angle = body side slip = tire side slip C = cornering stiffness
Fy = C
C0 & mV = C1 & r IZ
1 + mV 2
C2 I ZV
C1
f C mV + aCf r IZ
218
Tire Behavior
Fx = C s s F y = C
Linear for small slip angles Saturates at higher slip angles (vehicle slides) Varies with loading Corning stiffness depends on tire only Peak force changes with surface () and tire
Dugoff Model
7000 6000 5000 Lateral Force (N) 4000 3000 2000 1000 0 -60
Fz = 2 kN Fz = 4 kN Fz = 6 kN Fz = 8 kN
-50
-40
-10
-50
-40
-10
VEL GPS
v v VP = VA + rA / P
Side Slip at any point on the Vehicle: P V 1 Y P = tan V P X Tire Force & Cornering Stiffness
& Fx = m& x & = Fyf + Fyr cos( ) Fy = m& y && = aFyf bFyr cos( ) M z = I z
Fyf
f = r =
tire f tire r
C f =
, C r =
Fyr
, Cx =
Fx % slip
221
222
Without ESC II
With ESC II
223
Vehicle Estimation
Infinity G35:
GPS Velocity & Position 6 DOF IMU Steer angle Wheel Speed
224
Starfire/Beeline GPS
DATRON Velocity
Position & Velocity Course Heading & Roll Data Logging Real Time Analysis
On Board PC
CAN
225
Uses GPS and low cost inertial sensors with a vehicle test-bed
Estimation of Sideslip and Yaw Rate Estimation of Understeer gradient
226
100 %
20
Velocity (m/s)
Moderate Acceleration/Braking
Velocity (m/s)
20 15 10 5 0 0 20 0
% slip
Heavy Braking
Wheel Speed Sensor GPS
10
15
% slip
2 0 2 4 0 10 20 30 Time (sec) 40
227
Sideslip Estimation
Parking Lot
5 Steer Angle (deg) 0 -5 -10 -15 0 2 Slip Angle (deg) 0 -2 -4 0 5 10 15 20 25 30 Datron GPS/IMU Bicycle Model Slip Angle (deg) 35 40 Steer Angle (deg) 2 0 -2 -4 -6 0 3 2 1 0 -1 0 20 40 60 Time (s) 80 100 120 20 40 60 80 Datron 100 GPS/IMU Bicycle Model 120
Test Track
10
15
20 Time (s)
25
30
35
40
228
Front Tire
1
Rear Lateral Force, Fyr (N)
x 10
Rear Tire
0.5
0.5
0.5
0.5
1 10
10
1 10
10
Still in linear region Measurement does not rely on any tire model
GPS and Vehicle Dynamics Lab
20
20 0 100
r (deg/s)
20 0 5 10 15 20 25 30 100
r (deg/s)
10
15
20
25
30
100 0
10
15 Time (s)
20
25
30
Experimental
Measured Dugoff Fit -10 -5 0 5 Front Slip Angle (deg) 10 15
-10 -15 10
-10 -10
-5
10
231
-10 -10 10
-5
10
GPS DATRON
315
-10 -5
232
Asphalt
-20 0 5 10 15 20 25 30 35 40 45
50 0 -50 0
10
15
20 25 Time (s)
30
35
40
45
20
10 5
Gravel
-20 0 60 5 10 15 20 25
40 20 0 -20 0 5 10
Time (s)
10
Time (s)
15
20
25
233
Asphalt
-5
-5
-10 -10
-10 -25
-20
-15
10
15
10
10
Gravel
0 Filtered Experiment Experimental Dugoff Fit -40 -30 -20 -10 Rear Tire Slip Angle (deg) 0
-5 -35
-30
-25
Dynamics -20 GPS -15 and -10 Vehicle -5 0 5 Front Tire Slip Angle (deg)
Lab
-5
234
11.5 11
10.5
Asphalt
10
15
20 25 Time (s)
30
35
40
6.5 0
10
15
20 25 Time (s)
30
35
40
45
9 8
7 6 5 4 3 2 0 5 10
Gravel
6 5.5 5 4.5
Time (s)
10
Time (s)
15
20
25
235
Deduce the relative change in rolling resistance and correlate to fuel consumption
Fengine =
237
Leads to levels of confidence of performance when GPS is not available Leads to improved performance when GPS is not available
GPS and Vehicle Dynamics Lab 238
0 1 0 1 y= r 1 0 1 0 Bias gyro
2 0 Qw = 2 0 bias
[ y = [
y = rgyro
gps
gps
heading
for 2 antenna
239
241
Residuals from yaw rate and GPS for extreme cornering experiment with incorrect vehicle parameters
242
Sideslip and yaw rate estimate improved using modified tire cornering stiffness
Estimator residuals from parking lot test with modified tire cornering stiffness
243