Optical flow based navigation for a VTOL UAV
Bruno Hérissé
Tarek Hamel
CEA LIST Fontenay-Aux-Roses, France
[email protected]
UNSA - CNRS Sophia Antipolis, France
[email protected]
LIST – DTSI
11/10/09
1
Introduction
Purpose : Design of controllers for a VTOL UAV using inertial data and visual optical flow measurements.
IMU (Inertial Measurments Unit)
Camera (OF Measurments)
DTSI
SRI-LIST
11th October 2009
2
Outline
Optical Flow
Landing strategy
Terrain/Wall Following
Experiments on a quad-rotor UAV.
DTSI
SRI-LIST
11th October 2009
3
State Model of the UAV
The system is nonlinear and under actuated. T Definition of the used reference frames. I : Inertial frame attached to the earth B : body-fixed frame attached to the Γ vehicle at the center of mass. Vehicle state: ξ : position of the center of mass in I v : speed of the center of mass in I R : orientation matrix from B to I Ω : rotational velocity in B Dynamic Model used : Inputs: T : Global thrust in B Γ : Control torque in B Translational
F=f(T,R)
dynamics
Rotational Dynamics
High gain controller DTSI
SRI-LIST
Already controlled
11th October 2009
4
Forward optical Flow Optical Flow is the relative velocity of objects in the image plane T Γ
v0 d0 2
d0
DTSI
SRI-LIST
11th October 2009
5
Divergent Flow Optical Flow is the relative velocity of objects in the image plane T Γ
Time to contact Time to collision
DTSI
SRI-LIST
11th October 2009
6
Optical Flow : applications ?
Smooth vertical landing [Herisse-IROS'08]
Terrain/Wall following [Herisse-ICRA'09]
DTSI
SRI-LIST
11th October 2009
7
Vertical Landing
The control law
ensures the asymptotic convergence of d to
0
The UAV does not collide the obstacle
DTSI
SRI-LIST
11th October 2009
8
Terrain following
The control law
ensures the asymptotic convergence of d to d *
The UAV does not collide the obstacle
DTSI
SRI-LIST
11th October 2009
9
Terrain following: proof of stability
Proof:
DTSI
SRI-LIST
11th October 2009
10
Terrain following: robustness?
The forward velocity
can vary over time.
The target surface can be non-planar The optical flow is noisy The surface can be sloped…
DTSI
SRI-LIST
11th October 2009
11
Terrain following: sloped target
DTSI
SRI-LIST
11th October 2009
12
Terrain following: sloped target
DTSI
SRI-LIST
11th October 2009
13
Terrain following: disturbed dynamics
DTSI
SRI-LIST
11th October 2009
14
Terrain following: attractive domain
DTSI
SRI-LIST
11th October 2009
15
Wall following (2 cameras)
DTSI
SRI-LIST
11th October 2009
16
Wall following (2 cameras)
DTSI
SRI-LIST
11th October 2009
17
Experiment on a quad-rotor UAV
IMU (Inertial Measurments Unit)
Camera (OF Measurments)
DTSI
SRI-LIST
11th October 2009
18
Experimental setup
Camera: - Focus: 2.1mm
Drone: - Embeded attitude control -running at 166Hz
Numerical transmission: - Attitude - IMU data - 70 ms required
Textured terrain
DTSI
Analogical transmission: - 2.4GHz - 20Hz
Ground station: - OF Processing - 15Hz
SRI-LIST
11th October 2009
19
Experiment: Vertical Landing
DTSI
SRI-LIST
11th October 2009
20
Experiment: TF with 1camera
DTSI
SRI-LIST
11th October 2009
21
Experiment: TF with 2 cameras
DTSI
SRI-LIST
11th October 2009
22
Conclusion
Computation of the translational optical flow over a textured flat target plane. A robust strategy for navigation in cluttered environment Controllers prevent collisions with environment Control approach has been experimented on a quad-rotor UAV to demonstrate performance.
DTSI
SRI-LIST
11th October 2009
23