Vision-Based Relative Navigation for Proximity Operations
and On-Orbit Servicing Technologies
Alfredo Valverde, Guangcong Zhang, Fu-Jen Chu, Nuno Filipe, Michail Kontitsis, Panagiotis Tsiotras, Patricio Vela
Georgia Institute of Technology, Jet Propulsion Laboratory
Dual Quaternion Based Controller
Motivation
 On orbit servicing requires reliable and precise estimation of
the relative pose (i.e., attitude and position) between the
service and target satellite and the accurate and fail-safe
control of the service satellite.
 Unit Dual Quaternions provide a compact representation
of motion in SE(3).
Vision Based Attitude (3DoF) Stabilization
Vision (cooperative case)
Vision Measurements and
Estimates of Attitude
BoxLoG response
Input
Rate Gyro Measurements and
Estimates of Angular Velocity
BoxLoG
Data
Association
𝑞𝐵/𝐼
Pose Estimate
(Dual Quaternion)
1 𝐼
= 𝑞𝐵/𝐼 + 𝜖 𝑟𝐵/𝐼 𝑞𝐵/𝐼
2
ePnP
Average FPS
𝜖 ≠ 0 𝑎𝑛𝑑 𝜖 2 = 0
AprilTag
Our
approach
2.1
12
Attitude Tracking Error
 Combined Translational and Rotational kinematic and
dynamic equations are of the same form as their Rotation
only counterparts.
NASA Phoenix project
Attitude
Controller
(3DoF)
Autonomous Inspection
Unit Dual
Quaternion
Attitude &
Position
Controller
(5DoF)
Overview
Autonomous Spacecraft Testing of Robotic Operations in
Space (ASTROS)
Dual Quaternion Multiplicative Extended Kalman Filter
(DQ-EKF)
Relative Navigation
Rate Gyro / IMU
Experimental Setup
Air-bearing pads and
hemispherical joint for
frictionless translation and
rotation respectively.
Error Unit Dual Quaternion
Absolute
Navigation
Four (4) Variable Speed
Control Moment Gyros (VSCMG).
Absolute Navigation (5DoF) for a Docking Maneuver using
Thrusters
Planned vs Actual Motion
Maneuver Phases
Dual Quaternion
based Estimator
Dual Quaternion
based Controller
Actuators
(12 Thrusters, 4
VS-CMGs)
Twelve (12) thrusters.
Rate Gyro, 3-axis
Accelerometers.
Vision
+
−
−
𝑞𝐵/𝐼
= 𝑞𝐵/𝐼
∆ ∗ 𝛿𝑞𝐵/𝐼
Cooperative Vision based Relative Navigation using a known
pattern on the target satellite.
More compact (6 parameters)
compared to EKF based on true
unit quaternion (8 parameters)
Pose Estimation Error
Simulink Environment
Two on-board computers
(xPCTarget for control, PC104 for vision).
DQ-EKF performs well and
compares favorably to
additive EKFs
(SQV-AEKF and QV-AEKF)
Color CCD camera.
Attitude Tracking Error
Velocity Estimation Error
Position Tracking Error
Results
Attitude Tracking using VS-CMGs (3DoF)
Attitude
Attitude Tracking Error
Un-cooperative Vision based Relative Navigation relies on
extraction, and registration of features from previously unseen
scenes (in progress).
Box-LoG
Detector
TEMPLATE DESIGN © 2008
www.PosterPresentations.com
Replaced by
SIFT / SURF
Acknowledgement
This work was supported by the AFRL research award FA945313-C-0201