Tracking Drone Orientation with Multiple GPS Receivers Mahanth

Tracking Drone Orientation with Multiple GPS Receivers Mahanth Gowda, Justin Manweiler, Ashutosh Dhekne, Romit Roy Choudhury, Justin Weisz 1

Tremendous excitement with drones 2

Despite the excitement …

Why don’t we see drones flying around?

Several instances of crashes and failure … 5

Why do drones crash? 6

Drone Stabilization Accelerometer Gyroscope Compass Inertial Measurement Unit (IMU) crucial for tracking 3 D orientation stabilization, control

IMU can fail Accelerometer Gyroscope Compass Inertial Measurement Unit (IMU) crucial for tracking 3 D orientation stabilization, control However, rotor vibration and magnetic interference derails the sensors … causing correlated failures

Wouldn’t it be nicer to have an alternative to IMU? Drones already use GPS for localization Can we leverage GPS for orientation sensing as well? 9

We propose Safety. Net: A failsafe mechanism for IMU failures Basic Idea: Use multiple GPS to compute 3 D orientation

Use multiple GPS to compute 3 D orientation 11

Use multiple GPS to compute 3 D orientation 12

Use multiple GPS to compute 3 D orientation 13

Use multiple GPS to compute 3 D orientation 14

Use multiple GPS to compute 3 D orientation 15

Use multiple GPS to compute 3 D orientation Translate Relative Positions into 3 D Orientation 16

Use multiple GPS to compute 3 D orientation 2 degree orientation accuracy 2 cm location accuracy Basic GPS 3 m Absolute Position Error Opportunity: We do not care about absolute location, we only need relative Differential GPS 15 cm relative error Spatio-Temporal Information Fusion on small sized drone 17

Temporal Differentials Kalman Filter Spatial Differentials Background Carrier Phases Particle Filter Contributions Design of Differential GPS (15 cm) 18

Differential GPS (using carrier phase)

GPS Carrier Phase GPS Satellite Tru e Ran ge Pha se GP S C arri er W ave GPS Receiver Possible to measure precisely (2 -5 mm error) However, difficult to estimate N

GPS Carrier Phase GPS Satellite Tru e. R ang e GP S C arri Pha se er W ave GPS Receiver Real systems more difficult: However, since we only care about relative distances … Possible to eliminate some errors by subtracting phase equations across receivers, satellites, time, etc. For example …

Spatial Differentials S 1 Relative Position (Baseline) Clock Errors can be eliminated further 2 22

Temporal Differentials S Integer Ambiguity changes can be tracked using Doppler shift No Integer Ambiguity Change in orientation 23

Two Observations on Orientation Spatial Differentials Temporal Differential Drone Absolute Orientation, but polluted by Integer Ambiguity Change in Orientation, but unpolluted by Integer Ambiguity Analogous to compass and gyroscopes, but highly noisy We take a filtering approach to handle the dynamic errors 24

Bayesian Filtering Approach Temporal Differentials Transition q Orientation Measurement Spatial Differentials 25

Bayesian Filtering Approach Temporal Differentials Trans. q q q Meas. Spatial Differentials Integer Ambiguity ? 26

Integer Ambiguity Resolution depends on Flight Aggressiveness Two regimes of aggressiveness Low Dynamic Flights Kalman Filter Highly Aggressive flights Particle Filter 27

Orientation Estimation - Low Dynamic flights Trans. Static Drone Meas. Kalman Filter 28

Highly Aggressive Flights: Cycle Slips Trans. Cycle Slips Static Drone Incorrect Integer Ambiguity Meas. Kalman filter Kalman Filter derailed 29

How can we identify incorrect Integer Ambiguity (N)? 30

Identifying Poor Estimates of N Measurement Model 31

Correcting N via Particle Filters Trigger Particle Filter Back to Kalman Keep doing Kalman v v v ~~ 1 <1 High Conf. Low Conf. v v v v ~~ 1 . . . v v ~~ 1

Simple Particle Filter (Slow Convergence) Weigh and Propagate Initial Particles Measurement (q) 33

Adjusted Particle Filter (Accelerated Convergence) Propagate the Adjusted State Initial Particles Measurement (q) Combine Trans. And Meas. like Kalman Filter Provides near optimal combining for correct particles Accelerates convergence with few particles 34

Evaluation Platform Pi IMU GPS Go. Pro 35

Performance Vision Ground Truth GPS vs IMU 36

Safety. Net can track aggressive turns Tilt Orientation (degree) Sharp Angles (Truth) 37

Safety. Net can track aggressive turns 3 Orientation Angles Aggressive Regime 38

Overall Accuracy Tilt Angle Error (degree) Comparable to IMU Heading Error(degree) 39

Conclusion • Despite excitement, poor reliability is the show stopper for disruptive drone applications • We propose an orthogonal approach for drone state estimation using carrier phase GPS • We developed a Particle filter to systematically combine carrier phase data for orientation • Real Extensive flight tests achieve an accuracy comparable to IMU 40

Thank You Mahanth Gowda University of Illinois Urbana Champaign Systems & Networking Research Group (Sy. NRG) gowda 2@illinois. edu 41