Removing Drift from Inertial Navigation System Measurements RPUG Robert Binns Mechanical Engineering Vehicle Terrain Performance Lab

Transcription

1 Removing Drift from Inertial Navigation System Measurements RPUG 2009 Mechanical Engineering Vehicle Terrain Performance Lab December 10, 2009

2 Outline Laboratory Overview Vehicle Terrain Measurement System Addressing the Problem Correcting INS Drift in Terrain Measurements Summary Slide 2 2

3 Laboratory Overview Vehicles Passenger cars, commercial off-road, military vehicles, motorcycles (system level) Modeling and Simulation Terrain Modeling Measurement Performance Ride Handling Reliability Durability Slide 3 3

4 Vehicle Terrain Measurement System Scanning Laser Provides relative height measurement Inertial Navigation System DGPS: Establishes global coordinate system IMU: Mitigates low frequency body motion (<10 Hz) Accelerometers Mitigates high frequency body motion (>10 Hz) Slide 4 4

5 Addressing the Problem INS is capable of a establishing a geodetic (latitude & longitude) position with 2cm accuracy with differential GPS Experimentation shows artifacts of INS drift in height Max variation = +/- 10mm INS drift introduces run-to-run variation Negatively impacts IRI cross-correlation (~80%) ± 10mm ± 10mm Slide 5 5

6 Addressing the Problem Fundamental Question How do we recognize and remove error contributions from INS drift only? Slide 6 6

7 Correcting INS Drift Assumptions Neglect horizontal drift Elliptical height changes only in time Non-deformable terrain only INS treated as black box combining DGPS + IMU Slide 7 7

8 Correcting INS Drift Each measurement represents the sum of three sub-surfaces Measured Surface = Noise + INS Drift + True Surface Total Error Slide 8 8

9 Correcting INS Drift The Total error must be separated into INS drift and noise. Singular Value Decomposition determine contributions from different shapes to the error Noise must be zero-mean and is not correlated 0.3 to the INS drift Amplitude st Basis Vector 2nd Basis Vector Length of Vector space, n Slide 9 9

10 Correcting INS Drift Proof of Concept MnRoad Test Facility, Albertville, MN 160m section of asphalt 10 total measurements (alternating directions) Vehicle Velocity ~ 10 m/s 10mm Longitudinal Spacing 20mm Transverse Spacing Slide 10 10

11 Correcting INS Drift Drift Noise x 10-3 noise Drift (m) ±10mm Amplitude, [meters] ±1mm Longitudinal Distance, m Longitudinal Distance, [meters] Slide 11 11

12 Correcting INS Drift Proof of Concept Longitudinal view of terrain surface Terrain Height, [meters] No INS Drift Correction Profile Profile 2 Profile Profile 4 Profile Profile 6 Profile 7 Profile Profile 9 Profile Longitduinal Distance, [meters] Terrain Height, [meters] Basis Vectors Profile Profile Profile 3 Profile Profile 5 Profile Profile 7 Profile Profile 9 Profile 10 True Surface Longitudinal Distance, [meters] Slide 12 12

13 Summary INS Drift sufficiently characterized and removed Error contribution due to precision (noise) remains IRI cross-correlation improved to 98% Autocorrelation improved to 99% Future Work: Research effects on different road geometries (bank, grade, crowning, rutting) Slide 13 13

14 Large High-Resolution Display for Terrain Visualization RPUG 2009 Haeyong Chung Computer Science Vehicle Terrain Performance Lab December 10, 2009

15 Terrain Visualization Improve user s subjective understanding of the terrain data Have been used in various engineering simulations Industry is making use of terrain visualization in designing and planning Robert Haeyong Binns Chung Slide 15 15

16 VTMS Data Visualization A single dataset includes several millions of points. Difficult to visualize multiple VTMS datasets in small displays and desktop systems Massive amounts of computing power are required to render the high-fidelity terrain data. The resolution of the VTMS datasets is easily beyond the capability of current computer display and graphics systems. Robert Haeyong Binns Chung 16

17 Large High-Resolution Display (LHRD) Much higher DPI Wider field of view to terrain data Terrain visualization on a scale comparable to real life Terrain rendered in parallel Collaboration Robert Haeyong Binns Chung 17

18 Possible Applications for LHRD Visualization Data analysis of terrain data and simulation results Visualization of vehicular simulation Visual inspection of pavement health Planning and designing of roads Effective communication tool for stakeholders and public Robert Haeyong Binns Chung 18

19 Visualization Prototypes Robert Haeyong Binns Chung Slide 19 19

20 Visualization Prototypes Robert Haeyong Binns Chung Slide 20 20

21 Summary LHRD provides a novel platform to interactively visualize high-fidelity terrain data Terrain features can be displayed at a life-size scale Potential applications span both vehicle simulation and pavement communities Robert Haeyong Binns Chung Slide 21 21