Multi GNSS attitude estimation of UAVs during landing
Marton Farkas 1) 2) , Szabolcs Rozsa 2) , Balint Vanek 1)
marton.farkas@sztaki.mta.hu
Systems and Control Lab, Hungarian Academy of Sciences Institute for Computer Science and Control, Budapest, Hungary 1)
Department of Geodesy and Surveying, Faculty of Civil Engineering, Budapest University of Technology and Economics, Budapest, Hungary 2)
The project leading to these results has received funding from the European Union’s Horizon 2020 research and innovation programme, VISION, contract no. 690811.
Multi GNSS attitude estimation of UAVs during landing
Marton Farkas 1) 2) , Szabolcs Rozsa 2) , Balint Vanek 1)
marton.farkas@sztaki.mta.hu
Systems and Control Lab, Hungarian Academy of Sciences Institute for Computer Science and Control, Budapest, Hungary 1)
Department of Geodesy and Surveying, Faculty of Civil Engineering, Budapest University of Technology and Economics, Budapest, Hungary 2)
The project leading to these results has received funding from the European Union’s Horizon 2020 research and innovation programme, VISION, contract no. 690811.
Abstract
• Extended Kalman Filter (EKF) for estimating the baseline coordinates and single dif- ferenced integer ambiguities between two GNSS antennas and receivers [1]
• Single frequency, Multi GNSS (GPS, Glonass, Galileo) single baseline measurements
• Using code, phase and baselength measurements as the EKF’s inputs to determine the float solution
• Cycleslip detection based on triple differenced phase and the integrated doppler mesure- ments
• Integer ambiguity fixing
• Using modified LAMBDA (Least-squares AMBiguity Decorrelation Adjustment) method based on [2],[3]
• Inputs are the EKF’s states (baseline coordinates and transformed, double differenced integer ambiguities) the covariances, and the baselength between the antennas
• Searching for the best n integer ambiguity vector in the unconstrained space around the float solution and select the best vector in the baselength constrained space
• Validation with the norm of the fixed baseline coordinates
• Update the EKF’s baseline coordinate states and the covariances
• Computing bank (φ) or elevation (θ) and the heading (ψ) attitude angles from the single baseline coordinates
• Using surveying systems for validation, a small UAV and low-cost sensors for the testing
EKF
Baselength
Constrained LAMBDA ˆ
x, Qˆ
Validation |x˘b|
˘ x, Q˘
y
True
˘
xb, Q˘bb
x, Q x, Q
k-th epoch
(k + 1)-th epoch (k − 1)-th epoch
Fig. 1: Algorithm schematic State Vector and Covariance Matrix:
x =
xb xNT
, Q =
Qbb QbN
QN b QN N
Phase, Code and Baselength Measurement Vector:
y =
yph ypr yblT
b: Baseline components
N: Integer ambiguity components ˆ
v: Float solution variable
˘
v: Fix solution variable
Ground Test and Validation
Leica Total Station TOPCON HiPer II
GNSS receivers
Prism
Baselength = 1.5 m
Fig. 2: Ground Test
• Testing the algorithm under real, but ideal circumstances (clear sky, no disturbing terrain features) (Fig. 2)
• Reference angles were computed from the distances between the Prism and the Total Station
• Compare heading (ψ) and elevation (θ) angles from the GNSS (GPS, GLO) solution and the Total station’s solution
• Results (Fig. 3)
• Low mean and standard deviation values at the differences of the two kind of measurements
• Higher heading differences at the dynamic phases, probably time synchronization problem
0 40 80 120 ψ[◦ ]
Total Station GNSS Fix
218,000 218,400 218,800 219,200 219,600
−1 0 1 2
Mean: -0.02 Std: 0.23
∆ψ[◦ ]
·105
−1.5
−1
−0.5 0 θ[◦ ]
218,000 218,400 218,800 219,200 219,600
−0.5 0
0.5 Mean: 0.02 Std: 0.19
GPS SOW
∆θ[◦ ]
Fig. 3: Heading (ψ) and elevation (θ) angles,
and the differences of the two kind of measurements (∆ψ, ∆θ)
Flight Test
• Testing the algorithm with UAV flight data using low-cost sen- sors (Fig. 4)
• Reference angles were computed from the UAV’s IMU sensors (LIS331DLH accelerometer, L3G4200D gyroscope, HMC5883 magnetometer). Attitude angle’s accuracy (φ ± 5◦, θ ± 5◦, ψ ± 10◦) depends on the slideslip angle of the UAV.
• Compare heading (ψ) and bank (φ) angles from the GNSS (GPS, GLO, GAL) solution and the IMU solution
Tallysman TW4721 GNSS antennas
U-blox NEO-M8T Baselength = 2.5 m
Fig. 4: The UAV with the GNSS receivers and the antennas
Flight phases
• Controlled landing phases, with low glide angle
• Freestyle flight, with a barrel roll
250 300 350
Alt.[m]
Landing phases ←|→ Freestyle flight
468,400 468,500 468,600
6 10 14 18
GPS SOW
Satellitenumber GPS GPS+GLO GPS+GLO+GAL
Fig. 5: The flight altitude and satellite numbers
468,400 468,500 468,600
5 10 15 20
GPS SOW
[cycles]
Fig. 6: Mean absolute value of the phase triple differences
Results
• Fix solution rate is 74.5%, lower at the freestyle flight phase.
Float solution also has the trend with lower reliability.
• Higher angle differences at the higher dynamic phases, probably caused by slideslip flight, where the IMU solution accuracy is lower.
−100 0 100 ψ[◦ ]
IMU GNSS Fix GNSS Float
468,400 468,500 468,600
−20
−10 0 10
20 ∆ Fix Mean: -0.34 Std: 20.45 | ∆ Float Mean: -15.8 Std: 53.67
∆ψ[◦ ]
·105
−100 0 100
Barrel roll
φ[◦ ]
468,400 468,500 468,600
−10
−5 0 5
10 ∆ Fix Mean: -1.21 Std: 5.71 | ∆ Float Mean: -2.09 Std: 16.63
GPS SOW
∆φ[◦ ]
Fig. 7: Heading (ψ) and bank (φ) angles,
and the differences of the two kind of measurements (∆ψ, ∆φ)
Future plans
• Cycleslip determination and reconstruction using ac- celerometer sensor
• Tight fusion with low-cost IMU sensors for position and orientation estimation
• Validation with tactical grade sensors
The VISION project (Validation of Integrated Safety-enhanced Intelligent flight cON- trol) is an Europe/Japan collaborative research project. To enhance air transport safety, the main objective of VISION is to validate smarter technologies for aircraft Guidance, Navigation and Control (GNC) by including Vision-based systems, Advanced detection and resilient methods.
[1] Farkas, M. Short baseline static and kinematic GPS phase measurement analysis, using Unmanned Aerial Vehicle, Scientific Students’ Associations Conf., BUTE, 2015
[2] Giorgi, G. GNSS Carrier Phase-based Attitude Determination Estimation and Ap- plications. ISBN 978-94-6186-019-4
[3] Teunissen, P.J.G. Integer least-squares theory for the GNSS compass. J Geod (2010) 84: 433