You are on page 1of 6

Preprints of the 19th World Congress

The International Federation of Automatic Control
Cape Town, South Africa. August 24-29, 2014

Leader-follower Pose Consensus for
Heterogeneous Robot Networks with
Variable Time-Delays ?
Carlos I. Aldana ∗ , Emmanuel Nu˜
no ∗∗ , Luis Basa˜
nez ∗

Institute of Industrial and Control Engineering,
Technical University of Catalonia. Barcelona 08028, Spain.
{carlos.aldana, luis.basanez}@upc.edu.
∗∗
Department of Computer Science, CUCEI,
University of Guadalajara. Guadalajara 44430, Mexico.
emmanuel.nuno@cucei.udg.mx.
Abstract: This paper proposes a distributed proportional plus damping (P+d) control
algorithm to solve the leader-follower problem in which a network of heterogeneous robots,
modeled in the operational space, has to be regulated at a given constant leader pose (position
and orientation). The leader pose is only available to a certain set of followers. A singularity-free
representation, unit-quaternions, is used to describe the robots orientation and the network is
represented by an undirected and connected interconnection graph. Furthermore, it is shown
that the controller is robust to interconnection variable time-delays. Experiments, with a network
of two 6-Degrees-of-Freedom (DoF) robots, are presented to illustrate the performance of the
proposed scheme.
1. INTRODUCTION
The operational space is a subspace of the Special Euclidean space of dimension three, denoted SE(3). It is wellknown that the minimum number of coordinates required
to define the pose of an object in a three-dimensional
space is six: three for the position and three for the
orientation (attitude). Operational space control becomes
evident when cooperative tasks have to be described and
performed by multiple robot manipulators that could be
kinematically dissimilar (heterogeneous)(Liu and Chopra,
2012; Aldana et al., 2012). The practical applications of
multi-robot systems span different areas such as underwater and space exploration, hazardous environments (search
and rescue missions, military operations), and service
robotics (material handling, furniture assembly, etc.).
The consensus problem involves the design of algorithms
such that agents can reach an agreement on their states,
or on a common objective. This problem has been widely
studied for first and second-order linear time invariant
systems in the generalized coordinates space, that is the
case of linearized robots in the joint space. Some of the
proposed approaches include (Hu et al., 2013) where a
hybrid consensus control protocol is reported; (Sun, 2012)
that deals with uncertain topologies with interconnection
delays; (Abdessameud and Tayebi, 2013) that presents
a partial state feedback consensus controller; and (Nu˜
no
et al., 2013) that use a proportional plus damping (P+d)
control algorithm to solve the leaderless consensus problem in the joint space. For a comprehensive study and
? This work has been partially supported by: the Mexican CONACyT doctoral grant 168998; the Mexican CONACyT project CB129079; and the Spanish CICYT projects DPI2010-15446 and
DPI2011-22471.

Copyright © 2014 IFAC

further reference along this line, the reader may refer
to (Olfati-Saber and Murray, 2004). There are several
interesting papers dealing with Euler-Lagrange (EL) systems in the operational space, however most of them
only address the orientation part. Among these are (Liu
and Chopra, 2012) where an adaptive controller is proposed to solve the leader-follower attitude-only consensus
problem in networks of heterogeneous robot manipulators;
(Abdessameud et al., 2012) that proposes synchronization schemes to resolve the leader-follower and leaderless
problems for groups of rigid bodies in the presence of
communication delays; and (Ren, 2007) where three synchronization cases for the attitude alignment of spacecrafts
are presented.
Compared to the previously cited consensus solutions, this
paper proposes a leader-follower consensus algorithm for
networks of robots modeled in the operational space that
takes into account the full pose of the robots and not
only the attitude. This control scheme can be employed in
non-identical heterogeneous robots with different Degrees
of Freedom (DoF). The proposed controller is a simple
distributed P+d algorithm that is model independent (in
the sense that it is not necessary to know the robot’s
inertia matrix or the Coriolis forces) and is robust to
interconnecting variable time-delays. Furthermore, compare to other schemes that use singular minimum orientation representations, this work employs the singularity-free
unit-quaternions. Moreover, under the assumption that, at
least, one follower has a direct access to the leader pose,
it is shown that the proposed leader-follower controller
ensures the asymptotic convergence of the robots pose to
the constant leader pose and the asymptotic convergence
to zero of the linear and angular velocities. Finally, an
experimental validation of the leader-follower controller

6674

q ¨i ∈ R . Its EL-equation of motion. P2. 2. from the unit norm constraint. N }. yields q ¨i = J† (qi )υ˙ i + J˙ † (qi )υ i . ¯ i (qi . See Fig. 1992) for a detailed list of properties and operations involving unit-quaternions). 1 for the location of the reference frames and the pose vectors in an example scenario. q˙ i ) := J†i i i The operational space model (5) has the following wellknown properties (Spong et al. q˙ i )υ i | ≤ ηi |υ i |2 . ξ i ] . The matrix M P3. L2 -norm t≥0 R∞ as kf k2 := ( 0 |f (t)|2 dt)1/2 . R>0 := (0.. 2005): P1. is given by   p˙ i υi = = Ji (qi )q˙ i (2) ωi where υ i ∈ R6 and Ji (qi ) ∈ R6×ni is the geometric Jacobian matrix.. (6) Ki (υ i ) = υ > 2 i and Ui is the gravity potential energy such that gi (qi ) := ∂Ui ∂qi .. q˙ i )q˙ i + g ¯i (qi ) = τ i M qi + C (1) ni where qi . M where Mi (qi ) := J†i i i   >  ¯ i (qi )J˙ † + C ¯ i (qi . |x| stands for the standard Euclidean norm of vector x. where Ki is the kinetic energy given by 1 Mi (qi )υ i . Using the total energy function Ei (υ i . South Africa. The following notation is used throughout the paper. where N is the number The subscript i ∈ N of followers nodes of the network. is where i ∈ N given by ¯ i (qi )¨ ¯ i (qi .g. > ηi2 + β i β i = 1 (refer to (Chou. x ≡ x(t). M Ci (qi . 1k and 0k represent column vectors of size k with all entries equal to one and to zero. 6675 . respectively. 6 > > > where fi ∈ R . for j 6= i. respectively. ¯ := {1. q˙ i ) is a bounded operator. mi ∈ R3 represent the Cartesian forces and moments. veloc¯ i (qi ) ∈ Rni ×ni ities and accelerations.. d P4. q˙ i )J† .e.2 Orientation in the SE(3) A unit-quaternion ξ i ∈ S 3 can be split in two elements: one scalar term ηi ∈ R and one vectorial term β i ∈ R3 . respectively. qi ) = Ki (υ i ) + Ui (qi ). revolute joints robot manipulator and that the interconnection of the network can be represented using graph theory. 1 shows an example of a robot network with a followers network compose by three robots.1 Dynamics of the nodes Every ith-node is modeled as a ni -DoF robot manipulator. It is assumed that each follower node contains a fully-actuated. fi := [hi . ni 6= nj .. respectively. 2014 they can have different numbers of DoF. (4) i i Expressions (2). expressed also relative to a common reference frame. e. Fig. gi (qi ) := J† g ¯i (qi ). q˙ i )υ i + gi (qi ) = fi (5)  >  > ¯ i (qi )J† . The L∞ and L2 spaces are defined as the sets {f : R≥0 → Rn : kf k∞ < ∞} and {f : R≥0 → Rn : kf k2 < ∞}. R≥0 := [0. Note that since the robots are heterogeneous The relation between the joint velocities q˙ i and the linear p˙ i and angular ω i velocities of the ith-end-effector.19th IFAC World Congress Cape Town. the following relation between joint torque τ i and the Cartesian forces fi is obtained τ i = J> (3) i (qi )fi . e.g. The spectrum of the square matrix A is denoted by σ(A) where the minimum and the maximum of its spectrum is denoted by σmin (A) and σmax (A). Mi (qi ) is symmetric and there exist λmi . the identity and all-zeros matrices of size k × k. The pose of the ith-end-effector. ˙ i (qi ) − 2Ci (qi . and j ∈ N relative to a common reference frame. β > i ] and. is denoted by the vector xi ⊂ R7 and it contains the position vector pi ∈ R3 and the orientation unit-quaternion 1 ξ i ∈ S 3 . SYSTEM MODEL The complete dynamical description of the system contains two basic elements: i) the dynamics of the nodes and.. except for those which are time-delayed. λMi > 0 such that 0 < λmi I6 ≤ Mi (qi ) ≤ λMi I6 . respectively. The argument of all time dependent signals is omitted. If υ i . (3) and (4) allows to transform (1) to its corresponding operational space model. Pre-multiplying (2) by the Jacobian pseudo-inverse J†i (qi ) and differentiating. > Thus ξ i := [ηi . it can be shown that (5) represents a passive map from force fi to velocity υ i . Using the principle of the virtual work. ¯ . R := (−∞. 2. For all qi . M is the symmetric and positive definite inertia matrix. . g and τ i ∈ Rni is the torque exerted by the actuators (motors). such that > > xi := [p> i . A> (AA> )−1 is its Moore-Penrose pseudo-inverse matrix denoted by A† . q˙ i ) ∈ Rni ×ni is the Coriolis and centrifugal effects C matrix. given by Mi (qi )υ˙ i + Ci (qi . in joint space. respectively. ii) the interconnection of the nodes. there exists ηi ∈ R>0 such that |Ci (qi . q˙ i ) is skew-symmetric. ∞). υ˙ i ∈ L∞ then dt Ci (qi . mi ] and hi . Example of a robot network and the graph representing the interconnection using a network composed of two 6-DoF robots is presented. υ i . ∞).. i. 2. ¯ . For any function f : R≥0 → Rn . q˙ i . x(t − T (t)). 1 The set S 3 ⊂ R4 represents an unitary sphere of dimension three and it is defined as S 3 := {ξ ∈ R4 : |ξ|2 = 1}. are the joint positions. For any matrix A ∈ Rn×m . defined via the Christoffel symbols of the first ¯i (qi ) ∈ Rni is the gravitational torques vector kind. the L∞ -norm is defined as kf k∞ := sup |f (t)|. August 24-29. Ik and Øk represent. ∞). q˙ i . 1. Fig.

South Africa. relative to the world frame.19th IFAC World Congress Cape Town. August 24-29. between two different frames. 2014 The unit-quaternion ξ i can be easily obtained from the direct kinematics function of each robot manipulator. can be described by ˜ ij := Ri R> ∈ SO(3). via the rotation matrix Ri ∈ SO(3) := {Ri ∈ R3×3 : R> i Ri = I3 . The unitthe rotation matrix R j quaternion describing such orientation error is given by ξ˜ij = ξ i . The orientation error. det(Ri ) = 1} (Spurrier.. Spong et al. 1978. 2005). Σi and Σj .

ξ ∗j =  η˜ij ˜ij β  ξ> i ξj = ηj β i − ηi β j − S(β i )β j   η η + β> β = i j> i j −U (ξ i )ξ j   (7) A1. from the j-th robot to the i-th robot. If Assumptions A1 and A2 hold. the time–derivatives T˙ji (t) are bounded. 2004): a) L is symmetric and hence L> 1N = 0N .z> Lz = 12 > where . P Prank(L) = N 2−1. The interconnection graph is undirected and connected. is subject to a variable time-delay Tji (t) with a known upper-bound ∗ Tji . Further. and σm (L) = 0 has multiplicity one. The information exchange. and c) For aij (zi − zj ) ≥ 0. b) The spectrum of L satisfies σ(L) ≥ 0. any z ∈ RN . With regards to the interconnection time-delays it is assumed that A2. Hence. Additionally. it holds that 0 ≤ Tji (t) ≤ ∗ Tji < ∞. then the interconnection graph exhibits the following properties (Olfati-Saber and Murray.

The following lemma provides an interesting property of the composed Laplacian matrix L` := L + B that will be used in the proof of the main result and has been borrowed from Chapter 1. Assumptions A1 and A3 ensure that the leader pose is globally reachable from any of the N follower nodes.e. 3. culations show that β in turn. provided that the leader pose is only available to a certain set of followers (A3). S(·). A3. the following assumption is used in this paper: For any a. i. one bi` is strictly positive. used throughout the paper are: S(a)> = S(−a) = −S(a) and S(a)a = 03 . −β > (·) ] ¯ j∈Ni is the quaternion conjugate. b ∈ R3 .6 of (Cao and Ren.has the following assumption for these interconnections: quaternion and the angular velocity. |ξ i | = 1 then U(ξ i ) is a bounded operator. ˙ i ) = U(ξ˙i ). there exists some bi` > 0. For all ξ i ∈ S 3 and ξ˙i ∈ R4 . ... 2011) Consider a non-negative diagonal matrix B := diag(b1` . ¯. The control objective is to design distributed control laws fi . 2009) and in order to ensure that the interconnection forces are generated by the gradient of a potential function.3 Modeling the Interconnection The interconnection of the N agents is modeled using the Laplacian matrix L := [`ij ] ∈ RN ×N . bN ` ) ∈ RN ×N and suppose that. there exists at least one directed edge from the leader to any of the N followers.. there exists a path from the leader to any ith follower robot. it holds that > x˙ i = Φ (ξ i )υ i . a diagonal matrix B ∈ RN ×N is used −β i U(ξ i ) := . . (10) Using the normality condition. for all ξ i ∈ S . The following properties have been borrowed from (Fjellstad. some straightforward cal˜ij = 0 if and only if ξ i = ±ξ j . For all ξ i ∈ S 3 .  3. positive definite and of full rank. for all i ∈ N lim |υ i (t)| = 0. Similar to passivity-based (energy-shaping) synchronization (Sarlette et al. i. U> (ξ i )U(ξ i ) = I3 .. It is assumed that the interconnection graph fulfills Assumptions A1 and A2. 12 U> (ξ i )). Lemma 1. then the matrix L` = L + B is symmetric. LEADER-FOLLOWER CONSENSUS 2. U(ξ 3 P7. 2011). such that the network of N followers has to be regulated at a given constant leader > > 7 pose x` := [p> ` . 1994) and are used throughout the rest of the paper. denotes the quaternion product. Some well known properties of the skew-symmetric matrix operator. aij > 0 if j ∈ Ni and aij = 0 otherwise. At least one of the N follower robots has direct access to the leader’s constant pose x` . implies that U> (ξ i )ξ j = 03 .1 Problem Statement In this paper is considered a network of N kinematically different EL-systems in the operational space of the form (5). Hence.e. Hence. A key observation is that ξ i = ξ j and ξ i = −ξ j represent the same physical orientation. . S(a)b = a × b. P6. P5. in this work. (8) to model the leader-follower interconnections. ξ ∗(·) = [η(·) . lim xi (t) = x` . rank(U(ξ i )) = 3 and ker(U> (ξ i )) =span(ξ i ). This paper ηi I3 − S(β i ) The relation between the time-derivative of the unit. Lemma 1.2 Proposed Solution 2 The solution to the leader-follower consensus problem is established with the following operational space propor- 6676 . 2 Hence. (12) t→∞ t→∞ 3. ξ ` ] ⊂ R . at least. whose elements are defined as ( P aij i = j `ij = j∈Ni (11) −aij i 6= j where Ni is the set of agents transmitting information to the ith robot. . i. Since. (Cao and Ren. defining Φ(ξ i ) := diag(I3 . is given by 1 (9) ξ˙i = U(ξ i )ω i . S(·) is the skew-symmetric i∈N matrix operator 2 and U(ξ i ) is defined as The Laplacian matrix models the followers interconnection   > and. Assume that A1 holds.e. relative to the world reference frame. This.

with the fact that U> (ξ i )ξ i = 03 . then the closed-loop system (15) has the following two different ¯. otherwise. it ¯ j∈Ni i∈N is straightforward to show that X X  aij (x˙ i +x˙ j )> (xi −xj ) = x> (L> − L) ⊗ I7 x˙ = 0 ¯ j∈Ni i∈N j∈Ni > > > ¯ Defining ξ := [ξ > 1 . x` | and |xi − xj |. where B` and L` are defined in Lemma 1 and ⊗ is the standard Kronecker product. x˙ 1 ] ∈ R . since L1N = 0N . all i ∈ N At this point the main result of this work can be stated. q i X + ki aij Φ(ξ i )eij . . damping di satisfies ∗ 2 X Tji ¯. bi` > 0 if the leader pose x` is available to the ith robot manipulator and bi` = 0. . is given by eij = xi − xj (t − Tji (t)). . clearly satisfy υ i = 06 and bi` ei` + aij eij = 06 . . at Rt the equilibrium. The closed-loop system (15) exhibits the following (scaled) energy 1 bi` 1 X Vi := Ki (υ i ) + |xi − x` |2 + aij |xi − xj |2 (22) ki 2 4 j∈Ni where Ki has been defined in (6). . ξ j (t − Tji (t)) = ξ j − t−Tji (t) ξ˙j (σ)dσ and. (15) j∈Ni It should be mentioned that. is given by   X di 1 > V˙ i = − |υ i |2 − ( x ˙ − x ˙ ) (x − x ) aij x˙ > e − i j i j i ij ki 2 j∈Ni Z t X di = − |υ i |2 − x˙ j (σ)dσ− aij x˙ > i ki t−Tji (t) j∈Ni 1 X aij (x˙ i + x˙ j )> (xi − xj ) − 2 j∈Ni Adding and subtracting the term x> ˙ i and doing some i x algebra. at the Rt equilibrium. evaluated along (15). ∀i ∈ N (21) 2di > ki `ii αi + ki aji αj j∈Ni Furthermore. Rt Similarly. where the scalar ρ(·) is defined as ρ(·) := x> (·) x P P > using (11) and since aij (ρi − ρj ) = 1N Lρ = 0. the ¯ = 3N and P5 ensure that ξ i = ±ξ ` . ρN ]> ∈ RN and x := [x> . for fact that rank(U) ¯. (20) can be written as ¯ > [(B` ⊗ I4 )(ξ − (1N ⊗ ξ ` )) + (L ⊗ I4 )ξ] = 03N U > aij U (ξ i )ξ i to the j∈Ni previous equation. this last equations is equivalent to ¯ > (L` ⊗ I4 )(ξ − (1N ⊗ ξ ` )) = 03N U Finally. controller (13) solves the leader-follower consensus problem provided that. . Lemma 1 and the Kronecker product properties ensure that rank(L` ⊗ I3 ) = ¯ . . . thus. x> 1 . for any αi . U(ξN )) 4N ×3N ∈R . the equilibrium (17) is unstable and the equilibrium (16) is asymptotically stable everywhere except at (06 . . −ξ ` ). August 24-29. di ∈ R>0 are the controller gains. The possible P equilibria of (15). yields X bi` U> (ξ i )(ξ i − ξ ` ) + aij U> (ξ i )(ξ i − ξ j ) = 03 . since (L` ⊗ I4 ) is of full rank then the trivial solution ξ = 1N ⊗ ξ ` satisfies this equation. South Africa. although ξ i = ξ ` and ξ i = −ξ ` represent the same physical orientation. the closedloop system (15) has two possible equilibria. (20) > where ρ := [ρ1 . This last.(18) j∈Ni −bi` U> (ξ i )ξ ` − X aij U> (ξ i )ξ j (t − Tji (t)) = 03 . Adding the term bi` U (ξ i )ξ i + P Using L1N = 0N .19th IFAC World Congress Cape Town. αj > 0. 7N > > > x˙ := [x˙ 1 . for all i ∈ N (υ i .. Proposition 2. p` . together > > with p := [p> 1 . A2 and A3. . t−Tji (t) p˙ j (σ)dσ = 03 . Proposition 3. pi . −ξ ` ). . . (14) The closed-loop system (5) and (13) is h υ˙ i = −M−1 ˙ i )υ i +di υ i +ki bi` Φ(ξ i )(xi −x` ) i (qi ) Ci (qi . for any pair of robots (i. pN ] . Further. . |xi − ¯ and j ∈ Ni . for all i ∈ N Using (10). pi = p` is the only solution to 3N .(19) j∈Ni Rt Now. Thus. However. The interconnection error eij .P N] . Property (P2) and the fact that x˙ ` = 07 ensure that V˙ i . Now. Proposition 2 formally states this fact and Proposition 3 shows that ξ i = −ξ ` corresponds to an unstable equilibrium point. . .  Proof. allows to write (18) as (B` ⊗ I3 )(p − (1N ⊗ p` )) + (L ⊗ I3 )p = 03N and thus. ξ ` ) (16) and (υ i . ξ˙j (σ)dσ = 04 . Under Assumptions A1. . (19) can be t−Tji (t) written as X −bi` U> (ξ i )ξ ` − aij U> (ξ i )ξ j = 03 . . 2014 tional plus damping injection controller X fi = −ki bi` Φ(ξ i )(xi −x` )−ki aij Φ(ξ i )eij −di υ i +gi (qi ) j∈Ni (13) where ki . . p` . (17)  Proof. . which in turn implies that j∈Ni bi` (pi − p` ) + X aij (pi − pj (t − Tji (t))) = 03 . pj (t − Tji (t)) = pj − t−Tji (t) p˙ j (σ)dσ and. ξ N ] and U := diag(U(ξ1 ). Note that Vi is positive definite and radially unbounded with regards to υ i . yields > (x˙ i + x˙ j )> (xi − xj ) =x˙ > ˙ i − x˙ j ) + x> ˙i i (xi − xj ) − xi (x i x − x> ˙j j x > =x˙ > ˙ i − x˙ j ) + ρi − ρj i (xi − xj ) − xi (x ˙ (·) . for all i ∈ N (18). Hence. If Assumptions A1 and A3 hold. equilibrium points. ξ i ) = (06 . . p` .. ξ i ) = (06 . . j). taking V = Vi it ¯ i∈N holds that   Z t X di X  |υ i |2 + x˙ j (σ)dσ  aij x˙ > V˙ = − i k i t−Tji (t) ¯ i∈N j∈Ni 6677 j∈Ni . . pi . (L` ⊗I3 )(p−(1N ⊗p` )) = 03N .

All υ i . 4. Moreover. Additionally. Despite the initial poses differences and the variable time-delays in the interconnection. Hence. and the upper bounds of the induced variable time delays are: ∗ T21 =0. |xi − x` | and |xi − xj |. Furthermore. t→∞ The experimental setup is shown in the Fig. from (9). called PhanTorque 6Dof 3 . and it is composed of two robots with 6-DoF. d2 =1. t→∞ Boundedness of υ˙ i .   ··· . South Africa. it can be observed that pose errors asymptotically converge to zero. d1 =1. the closed-loop system (15) has only two t→∞ possible equilibrium points. The communications between Simulink and the robots make use of a custom library.6. i. from (9). b1` = 1 and b2` = 0. x˙ i ∈ L2 implies that υ i ∈ L2 . in the same spirit as in (Nu˜ no et al. (υ i .19s and ∗ T12 =0. together with Property (P4).5 High Force and one PHANTOM Premium R 1. for ¯ . 2. pi . yields  Z t X di Z t 2 2 ˙ |ξ i (θ)| dθ |x˙ i (θ)| dθ + 3 V(t) − V(0) = − k 0 0 ¯ i i∈N Z t Z θ X X − aij x˙ > x˙ j (σ)dσdθ. 2013). recalling that `ii = + X X Fig.. Barbˇ alat’s Lemma with υ i ∈ L∞ ∩ L2 and υ˙ i ∈ L∞ supports the fact that lim υ i (t) = 06 . it is possible to integrate V˙ from 0 to t and then apply Lemma 1 in (Nu˜ no et al. These robots are commercially availR able from Geomagic .5 with their control computers interconnected through a local area network. ξ i ) = (06 . 2014 Since V does not qualify as a Lyapunov Function. thus ¯ := {1. defined as µ := Ψ> 1N .. 5. This last implies that ¯ and j ∈ Ni . 3 and the Fig. kx˙ N k22 ≤ V(0) N Ψ kx ki ¯ i∈N where     Ψ=   a ∗T 2 `11 α1 − 122α1 21 2 a ∗T 2 − 212α2 12 kd22 − `222α2 d1 k1 − . . 2}. V(t) ≤ V(0).    . V˙ < 0. x˙ i ∈ L2 and V ∈ L∞ ..35 ensures that (21) holds. |xi − xj | ∈ L∞ . t→∞ 3 The libraries are publicly available at http://sir.php/Projects/PhanTorqueLibraries 6678 . The gravitational torque vectors and the estimated values of the two Premium devices.. 2013). 2.4. The data transmission of the robots’ pose is done through UDP ports and. such that µi > 0. EXPERIMENTAL VALIDATION i∈N j∈Ni which can be further written as X 3di  > V(t) + kξ˙i k22 + 1> ˙ 1 k22 . − − aij k ` 2 i ii ¯ V(t) − V(0) ≤ − 4. namely (16) and (17).. i∈N ∗ ! 2 Tji αi kx˙ i k22 + kx˙ j k22 . imply that υ˙ i is uniformly continuous. the proportional and damping injection gains are fixed to: k1 =10. aij j∈Ni ∗ 2 X 3di X X Tji kξ˙i k22 + kx˙ j k22 aij k 2α i j ¯ ¯ j∈Ni i∈N   i∈N X X αi di kx˙ i k22 . . Experimental validation setup. it can be inferred that 4|ξ˙i |2 = |ω i |2 thus. using |υ i |2 = |x˙ i |2 + 3|ξ˙i |2 .. if di is set according to (21) then there exists µ ∈ RN . .edu/wikis/ roblab/index. from the closed-loop system (15). for all i ∈ N these bounded signals together with Property (P3) and the fact that |ξ i | = 1 ensure. α2 =0. Using (22) it is shown that (16) corresponds to a minimum energy point and since V(t) is a decreasing function. i. The resulting leaderN follower interconnection can be seen in the Fig. 2006). . p` . These experimental results verify the theoretical results of Proposition 3. (θ) i 0 ¯ j∈Ni i∈N θ−Tji (θ) ¯. The Jacobian matrix is detailed in (Rodriguez and Basa˜ nez. from Fig.e.e. which in t→∞ turn implies from (9) that lim x˙ i (t) = 07 . k2 =7. 2009) to the double integral terms. lim t→∞ 0 then lim υ˙ i (t) = 06 . |xi − x` |. August 24-29.. . Furthermore. υ i .19th IFAC World Congress Cape Town. both robot poses (x1 and x2 ) asymptotically reach the leader pose (x` ) as can be seen In the Fig. and. Hence all i ∈ N  X  3di 2 2 ˙ V(t) + kξ k + µi kx˙ i k2 ≤ V(0). one PHANTOM R Premium 1. that υ˙ ∈ L∞ . 2005). .. ki i 2 ¯ i∈N Therefore. and it also returns the transformation matrix of the robots end-effector. The interconnection weights are set to: a12 = a21 = 0.upc.25. 2. 2 2α j ¯ j∈Ni i∈N P aij then it holds that Moreover. a couple of artificially induced delays have been added (Salvo Rossi et al. can be found in (Aldana et al.3. to emulate a transcontinental communication. Setting α1 =0. . for any αi > 0 and i ∈ N  X di  kx˙ i k22 + 3kξ˙i k22 V(t) − V(0) ≤ − k ¯ i Invoking Proposition 2 it holds that.11 and Simulink version 7. dN kN − `N N αN 2 Clearly.6. ξ ` ) is asymptotically stable everywhere except at the unstable equilibrium point (17). . The two PHANTOM robots define the followers network and the leader constant pose is sent only to Node 1. − ∗ 2 aN 1 T1N 2αN − ∗ 2 aN 2 T2N 2αN 2 a1N ∗ TN 1 2α1 ∗ 2 a T − 2N2α2 N 2 ··· −  ··· ..25s. since Z t υ˙ i (σ)dσ = lim υ i (t) − υ i (0) = −υ i (0). This concludes the proof. The controllers have been programmed using Matlab version 7. when lim υ i (t) = t→∞ lim υ˙ i (t) = 06 . any perturbation from (17) will drive the system to (16). . This library allows to directly set the torques of the 6-DoF in the PHANTOM Premium models..

2014 [m] 0. R. 2006. K. may 2012. 8 (1):53–64. Chou.ℓ py. and M.05 0 0 2 4 6 time [s] 8 10 Fig. 2009. Technical University of Catalonia. 28(7):895–910. South Africa. J. Tayebi.2 py.04 0.2 0. W. Ortega. and L. H.upc. Sepulchre. Proc. it is shown that the controller is robust to interconnection variable time-delays. I. 2013. feb. Sept. Aldana. F. Nu˜ no. Control of bilateral teleoperators in the operational space without velocity measurements. A. R. Group consensus in multi-agent systems with hybrid protocol. 4. 28(1):268–275. Basa˜ nez. Murray.5/6DOF (http://upcommons. Fjellstad.edu/pfc/handle/2099. 26(6):1503– 1508. C. 1 η1 0. L. Cao and W. Conf.9 0.1 β3. 1978.ℓ β2. Robot Modeling and Control. Spurrier. Distributed attitude alignment in spacecraft formation flying.02 0 px. Contr. Song. Automatica. Sun.15 [m] 0. IEEE Trans. with a simple decentralized proportional plus damping controller and sufficiently large damping the network asymptotically reach a consensus pose at the given leader pose. 349(3):1061–1073.ℓ −0. Yu. On consensus algorithms design for double integrator dynamics. 49(9):1520–1533. and H. 0.85 0 −0. A. A. W. IEEE Trans. E. J. Palmieri. Sarras. Ren. Signal Process. L..1/14587)..2 0. IEEE Trans. E. Robot. IEEE Trans. Robot. IEEE Trans. Consensus in networks of nonidentical Euler-Lagrange systems using P+d controllers. J. It is proved that.05 0 0 2 4 6 time [s] 8 10 Fig.1 0. 1994. Comment on singularity-free extraction of a quaternion from a direction-cosine matrix. are shown to support the theoretical results of this paper. 2005. 57(9): 2405–2411. Basa˜ nez.. and L. J. 2012. Quaternion kinematic and dynamic differential equations.. A. Leonard. Institute of Industrial and Control Engineering.. Both. Iannello. E. and G.3 0.-X. 45(2): 572–577.C.1 0. Franklin Inst. Controlled synchronization of heterogeneous robotic manipulators in the task space. Spacecraft and Rockets. 2012. Adapt. Aldana. Vidyasagar. Franklin Inst. Nu˜ no.2 |e12 | 0.2 β3. Control. Conf. A. C. Y. the dynamics and the controller are defined in the operational space using unit quaternions to represent the orientation.. Polushin. Sign. with a network of two 6-DoF robot manipulators. Experiments. Control of unmanned underwater vehicles in six degrees of freedom: a quaternion feedback approach.15 0. Int. Spong. University of Trondheim. Basa˜ nez. 2007. J.1 py. E. In IEEE/RSJ Int.4 0 −0. Hutchinson. I. Basa˜ nez.ℓ −0. ISBN 978-0-47164990-8. 2013. Sarlette.-A. Zhang. 2005.2 β2. 6679 .. O. Technical report. Nu˜ no.1 px.1 0. Hu. P.1 pz. pages 4274 –4279. G. Autom.M. and L. Models. 54(2):530–541. J. Pose errors. Furthermore. July 2009. on Intell. 2012.5 |e1ℓ | 0. Distributed Coordination of Multiagent Networks: Emergent Problems. Wiley. 2011. Robots’ orientation (Unit-quaternions). Olfati-Saber and R. Average consensus in networks of dynamic agents with uncertain topologies and time-varying delays. In IEEE Int.2 β1. E.19th IFAC World Congress Cape Town.2 pz.95 η2 ηℓ 0.2 β1. Attitude synchronization of multiple rigid bodies with communication delays. Spong. Salvo Rossi.-G. Y. Future research avenues span the development of adaptive controllers for the estimation of the gravity terms. W. Basa˜ nez. 5. 2013. Position tracking for nonlinear teleoperators with variable time-delay. R. 3. 350(3):575–597. Robots and Syst. C. W. 49 (1):253–260.ℓ pz. Liu and N. Y. Abdessameud. S. E. R. Ren.1 0 0. Int. Joint end-to-end loss-delay hidden markov model for periodic udp traffic over the internet. Rodriguez and L.1 β1. Autom.2 px. 1992. Consensus problems in networks of agents with switching topology and timedelays. Control.02 [m] 0. Res. Robot. Abdessameud and A. M. Autonomous rigid body attitude synchronization. August 24-29.2 0. Bilateral teleoperation of cooperative manipulators. PhD thesis. and M. and N. 5. Robot. Springer-Verlag. and Issues.. Norwegian Institute of Technology.1 0. Modelo cinem´atico de la interfase h´aptica PHANToM 1. Autom.05 0 0 2 4 time [s] 6 8 10 Fig. 15(4):255. Autom. Chopra.1 β2. Tayebi. IEEE Transactions on Robotics. 21(2-3):95–113. Romano. CONCLUSION This paper presents a controller that solves the leaderfollower operational space consensus problem in networks composed of multiple robot manipulators. REFERENCES A.ℓ β3. and I.4 0. Nu˜ no. 2004.. Robots’ position. Automatica. 2013.