• Aucun résultat trouvé

M.LeGallic,J.Tillet,F.LeBars,L.Jaulin TightSlalomControlforSailboatRobots

N/A
N/A
Protected

Academic year: 2021

Partager "M.LeGallic,J.Tillet,F.LeBars,L.Jaulin TightSlalomControlforSailboatRobots"

Copied!
24
0
0

Texte intégral

(1)

Tight Slalom Control for Sailboat Robots

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin

IRSC 2018, Southampton September 1, 2018

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(2)

Tight slalom

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(3)

wind

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(4)

We consider a mobile robot [1]

_ x = f ( x , u ) p = g ( x )

with an input vector u = (u 1 , . . . , u m ) and a pose vector p = (p 1 , . . . ,p m+ 1 ) .

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(5)

Control m + 1 state variables and not only m of them.

Perform a path following instead of a trajectory tracking.

The controller makes p collinear (instead of equal) to the required ˙ eld.

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(6)

Method

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(7)

Dubins car: 

˙

x 1 = cos x 3

˙

x 2 = sin x 3

˙

x 3 = u

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(8)

Output

e = x 3 + atanx 2 . Thus

˙

e = ˙ x 3 + x ˙ 2

1 + x 2 2 = u + sin x 3 1 + x 2 2 . We want

˙

e + e = 0.

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(9)

We get:

u = −x 3 − atan x 2sin x

3

1 +x

22

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(10)

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(11)

Generalization

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(12)

We want to follow the eld ψ( p ) .

This can be translated into ϕ (ψ (p), p) = ˙ 0, where ϕ ( r , s ) = 0 ⇔ ∃λ > 0 , λ r = s .

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(13)

We dene

e = ϕ (ψ ( p ), _ p ) = ϕ

ψ( g ( x )), ∂ g

∂ x ( x ) · f ( x , u )

. Since dime = dimu = m , we can make e → 0.

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(14)

Van der Pol cycle

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(15)

The car has to follow the limit cycle of the Van der Pol equation:

ψ ( p ) =

p 2

− 0 . 01 p 1 2 − 1 p 2 − p 1

.

We take g ( x ) = (x 1 ,x 2 ) T to build paths in the (x 1 ,x 2 ) -space.

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(16)

We get that nal controller is u = − sawtooth

x 3 − atan2

x

2

100

1

− 1

x 2 − x 1 ,x 2

+

x2

1001

− 1

x

2

+x

1

· sin x

3

+x

2

·

x1x2 cosx3

50

+

x2

1001

− 1

sin x

3

+ cos x

3

x

22

+

x2

1001

− 1

x

2

+x

1

2

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(17)

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(18)

Controller

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(19)

We consider sailboat [2]:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

˙

x 1 = v cos θ

˙

x 2 = v sin θ θ ˙ = −ρ 2 v sin2 u 1

˙

v = ρ 3 k w ap k sin (δ s − ψ ap ) sin δ s − ρ 1 v 2 σ = cosψ ap + cos u 2

δ s =

π + ψ ap if σ ≤ 0

− sign ( sin ψ ap ) ·u 2 otherwise w ap =

−a sin (θ) − v

−a cos (θ)

ψ ap = angle w ap

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(20)

The path to follow is

e ( p ) = 10sin p 1 10

− p 2 = 0 where e (p) is the error.

We want e ˙ = − 0 . 1 e .

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(21)

Thus

cos p 1 10

˙ p 1 − p ˙ 2

| {z }

˙ e( p )

= − 1 10

10sin p 1 10

− p 2

| {z }

e( p )

We take p ˙ 1 = 1, to go to the right. Thus:

ψ ( p ) = p ˙ 1

˙ p 2

=

1

cos p 10

1

+ 10 1 10sin p 10

1

− p 2

(1) which is attracted by the curve p 2 = 10sin p 10

1

.

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(22)

The controller is

u 1 = − 1 2 arcsin tanh

ω ˆ ρ

2

v

ω ˆ = −( sawtooth( θ − atan2 (b, 1 )) + 1 +b b ˙

2

b ˙ = 10 1 cos x 3 · cos x 10

1

− sin 10 x

1

10 1 sin x 3 b = cos x 10

1

+ sin x 10

1

10 1 x 2

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(23)

wind

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

(24)

L. Jaulin.

Automation for Robotics.

ISTE editions, 2015.

L. Jaulin and F. Le Bars.

An Interval Approach for Stability Analysis; Application to Sailboat Robotics.

IEEE Transaction on Robotics, 27(5), 2012.

M. Le Gallic, J. Tillet, F. Le Bars, L. Jaulin Tight Slalom Control for Sailboat Robots

Références

Documents relatifs

From what preceeds, it can be observed that any robot having a simple open-tree structure may be described by a sequence of n letters (n being the number of joints) R and P which

The “trajectory” toolbox from the RobOptim application layer is currently providing trajectores defined as B-Splines and associated mathematical functions to realize minimal

In this paper we present the HATP (Hierarchical Agent- based Task Planner) planning framework which extends the traditional HTN planning domain representation and seman- tics by

In what follows, we explain the selection of the performance criteria used to study the Parkour motion, which allowed us to identify the tasks for generating the movement using

In the case of robotics specifically, does humans’ perception of movement impact the way that they talk about moving machines.. At contrary, does natural language impact

C ongratulations to the IEEE Robotics and Automation Society (RAS) members recently elevated to Senior Member status by the IEEE Ad ­ mission and Advancement Senior Member

As a consequence, since the wrapper model makes it possible to build a representation to learn from, it is a perfect candidate for implementing Perceptual Learning inspired operators

In our work we have targeted these gaps in the software development for agent oriented cloud robotic sys- tems which will be found of use when imminently cloud robotics and