Biomimetic Algorithms for Coordinated

Motion: Theory and Implementation

Udit Halder1 and Biswadip Dey2:̊§;
arXiv:1503.04894v1 [cs.RO] 17 Mar 2015

March 18, 2015

Abstract
Drawing inspiration from flight behavior in biological settings (e.g. territorial bat-
tles in dragonflies, and flocking in starlings), this paper demonstrates two strategies
for coverage and flocking. Using earlier theoretical studies on mutual motion camou-
flage, an appropriate steering control law for area coverage has been implemented in a
laboratory test-bed equipped with wheeled mobile robots and a Vicon high speed mo-
tion capture system. The same test-bed is also used to demonstrate another strategy
(based on local information), termed topological velocity alignment, which serves to
make agents move in the same direction. The present work illustrates the applicability
of biological inspiration in the design of multi-agent robotic collectives.

Index Terms - Path Planning for Multiple Mobile Robots or Agents, Autonomous Naviga-
tion, Distributed Robot Systems, Wheeled Robots

1 Introduction
The last few decades have witnessed an increase in research efforts towards uncovering mech-
anisms behind pursuit behavior [2–5] and collective motion [6–9] in nature. Parallel to this
stream of endeavors, a number of mathematical models and appropriate feedback mecha-
nisms [10–18] have been introduced to bring these ideas from natural settings to unmanned
vehicular systems. Moreover, some groups in the robotics community have performed suc-
cessful implementation of these concepts [19, 20], and thereby demonstrated the power of a
bio-inspired approach towards synthesizing collective motion. This current work of ours is
similar in spirit, and provides an indoor demonstration of two strategies for coverage and
flocking behavior.
˚
This paper is an extended version of [1].
:
This research was supported in part by the Air Force Office of Scientific Research under AFOSR Grant
FA9550-10-1-0250. Experimental evaluation of the control laws was carried out in the Intelligent Servosys-
tems Laboratory, on a physical test-bed system for synthesis of collective behavior from fundamental building
blocks, supported by a FY2012 DURIP Grant from the AFOSR (FA2386-12-1-3002).
;1
Udit Halder is with the Department of Electrical and Computer Engineering, University of Maryland,
College Park, MD, USA. [email protected]
§2
Biswadip Dey is with the Department of Electrical and Computer Engineering, Institute for Systems
Research, University of Maryland, College Park, MD, USA. [email protected]
Figure 1: Mobile robot based experimental platform (Pioneer 3 DX) with two-wheel differ-
ential and caster.

In this paper, we have implemented mutual motion camouflage (MMC) [13] on a mobile
robot test-bed. Existing literature on dragonflies [21] provides qualitative analysis of territo-
rial battles, wherein the trajectories display spiraling motion consistent with the theoretical
predictions [22]. This particular bio-inspired control algorithm inherits an appealing cover-
age property through the mechanism of space filling curves, and our implementations are
able to reproduce coverage patterns similar to the predicted ones.
Although there has been a long history of control algorithms for flocking, almost every
model of collective motion predicts diffusive transport of information. But, contrary to the
existing models, recent findings [18] from starling flocks suggest that directional information
within a flock propagates with an almost constant speed, and this linear growth of infor-
mation can be explained by models with wave-like aspects. In the later part of this paper,
we have introduced a control strategy, called topological velocity alignment (TVA), which
conforms to this criterion and can explain how information about local neighbors can influ-
ence the agents in a flock to align their headings in a single common direction. Moreover,
an ongoing research work has shown that empirically observed curvature values in starling
flocks carry a strong resemblance with the curvature values predicted by the steering control
laws associated with TVA. Hence it seems reasonable to use TVA strategy for collective
motion synthesis. Furthermore, our implementation results in real robots have shown that
reduction in neighborhood size and external perturbation (similar to predator attack) can
split a flock into smaller subgroups.

2 Self-steering Particle Model

By considering an agent to be a unit-mass, self-steering particle, we use natural Frenet frame
equations [23] to describe its motion. This approach to describe an agent’s motion is based
on the fact that, while the unit tangent vector xi for a given trajectory ri is unique, we can
choose two unit vectors pyi , zi q such that txi , yi , zi u defines a right-handed orthogonal frame.
The evolution of this frame along the agent’s trajectory is given by

r9 i “ νi xi
x9 i “ νi pui yi ` vi zi q
(1)
y9 i “ ´νi ui xi
z9 i “ ´νi vi xi ,
2
where νi represents speed and (ui ,vi ) are the natural curvatures. In a similar way, by re-
stricting the motion of an agent to a planar setting, we can model its motion as

r9 i “ νi xi
x9 i “ νi ui yi (2)
y9 i “ ´νi ui xi ,

where xi denotes normalized velocity of the i-th agent and yi is the orthogonal rotation of
xi in the counter-clockwise direction. It should be noted here that this way of modeling a
trajectory requires only twice-differentiability of the trajectory.

Remark 2.1. As we use Hilare type [24] mobile robots (Fig 1) as a test-bed for control
algorithms, (2) provides a natural choice to model their motion.

3 Mutual Motion Camouflage (MMC)

Here we consider the mutual interaction between two agents each applying the same pursuit
law, while perceiving the other one as a target. As the dynamics of MMC in a planar
setting has been studied earlier [13], we just reiterate some key results in order to have
a comprehensive framework. Allowing different speeds for the agents, we begin with the
following symmetry:
u1 ν1 “ u2 ν2 “ u. (3)
Then the dynamics of the relative motion vectors, namely r “ r1 ´ r2 , g “ r9 “ ν1 x1 ´ ν2 x2
and h “ gK “ r9 K , can be expressed as

r9 “ g
g9 “ uh (4)
h9 “ ´ug.

Now we introduce three scalar shape variables defined as ρ “ |r|, γ “ pr ¨ gq{|r| and λ “
pr ¨ hq{|r|. Then, according to [10], we have
ˆ ˙ ˆ ˙
r K r
u “ ´µ ¨ r9 “ ´µ ¨ h “ ´µλ, (5)
|r| |r|

where, µ ą 0 denotes the feedback gain. As shown earlier, the dynamics of relative motion
(4) can be reduced to yield a second order dynamics given by

ρ9 “ γ
(6)
γ9 “ p1{ρ ´ µq δ 2 ´ γ 2 ,
` ˘

where, δ “ |g| “ |h| is conserved along any trajectory of (4). As detailed in the original work
[13], individual trajectories can be reconstructed from the solutions of (6). Moreover, the
solutions of the reduced dynamics (6) constitute level sets for another conserved quantity,
defined as
Epρ, γq “ ρ2 pδ 2 ´ γ 2 qe´2µρ “ Epρ0 , γ0 q. (7)

3
3.1 Stabilization through Dissipation
As it will be described in section 5, implementation of the original MMC law (5) in a labo-
ratory environment did not result in a satisfactory performance. The trajectories diverged
from the theoretical predictions, and the errors kept building up (Fig 4). This type of be-
havior can be attributed to the lack of damping in the reduced dynamics (6). As a result,
every small error was kept unchecked, and accumulation of these errors gave rise to a poor
performance of the MMC feedback law. Based on this observation, we considered a modified
version of the feedback law which introduced a dissipation term. As Epρ, γq was not staying
constant during an implementation of MMC feedback law (5), we added a dissipative term
in the feedback control to counter-act any deviation from the predicted trajectories. The
resultant control law can be expressed as
` ˘
utot “ u ` udis “ ´µλ ` kd λγ Epρ, γq ´ Ed , (8)

where Ed is set as the initial value of the conserved quantity Epρ, γq (Ed “ Epρ0 , γ0 q). Earlier
research [12] has shown that the modified control law (8) with kd ą 0 makes the periodic
orbit (with energy Ed ) orbitally asymptotically stable, and the corresponding domain of
attraction is characterized by tpρ, γq : ρ ą 0, ´δ ă γ ă δ, pρ, γq ‰ p1{µ, 0qu.

4 Topological Velocity Alignment (TVA)

Here we formalize the strategy of topological velocity alignment (TVA), and assume that each
member in a group of n-agents uses this strategy to move together while keeping its heading
parallel to the neighborhood center of mass velocity. Letting Ni denote the neighborhood of
the i-th agent, the center of mass (COM) velocity of this neighborhood is given by
1 ÿ
vCOM “ νj xj , (9)
|Ni | jPN
i

where |Ni | represents the number of neighbors influencing the i-th agent. Next, by assuming
that vCOM does not vanish to zero, we define the direction of the center of mass motion as
vCOM
xNi “ . (10)
|vCOM |

It should be noted that xNi is not well-defined over a thin set in the state space. As the
chance of getting into the thin set is essentially zero, we can overlook this situation for all
practical purposes. Now we introduce a contrast function
1
Θi “ pxNi ´ xi q ¨ pxNi ´ xi q “ 1 ´ xi ¨ xNi , (11)
2
as a quantitative measure for the misalignment between the heading of an agent and the
direction of motion of its neighborhood center of mass. Clearly, this contrast function (Θi )
assumes its minimum value (“ 0) whenever the i-th agent’s velocity is aligned with its
neighborhood center of mass velocity, and increases monotonically with increase in the mis-
alignment between them. Thus, Θi can be interpreted as a measure of departure from our
goal of achieving alignment.

4
 Next, by assuming a non-zero velocity for the neighborhood center of mass (vCOM ‰ 0),
we propose a control law ˆ ˙
xNi ¨ yi
ui “ µ
νi
ˆ ˙ (12)
xNi ¨ zi
vi “ µ ,
νi
where µ ą 0 denotes a positive gain, and yi , zi carry their usual meaning. Alternatively,
lateral acceleration for this choice of control laws (12) can be expressed as

alat
` “ ‰ ˘
i “ µν i xN i
´ xN i
¨ xi xi , (13)

and this provides a physical intuition behind (12) as the lateral acceleration is proportional
to the projection of the normalized velocity of its neighborhood center of mass onto the
transverse of its own direction of motion.

Remark 4.1. Earlier works [17, 25] have considered a very similar form of this control law
with three components for attraction (while the agents are far away), repulsion (to avoid
collision) and velocity alignment. However, the control law introduced in this current study
considers only velocity alignment, but extends the scope from a planar setting to a three
dimensional environment. Moreover it relaxes the assumption on uniform speed of the col-
lective by allowing the agents non-uniform and time-varying speed profiles. This relaxation
plays an important role in the context of applying this control law to a group of heterogeneous
agents.

4.1 Special Case: A planar 2-agent system

As analysis of an n-agent system with neighborhood defined as the set of K-nearest neighbors
poses hard challenges, we begin by considering a special case, namely a two-agent system
with the motion restricted to a planar setting. Clearly, the neighborhood center of mass
never loses speed in this case (as the neighborhood is nothing but the other agent).

4.1.1 State space and its reduction onto the shape space
By modeling the dynamics of an agent using natural Frenet frame equations restricted to a
planar setting (2), its position and the frame vectors can be packed inside a 3 ˆ 3 matrix
gi P SEp2q, the Lie group of rigid motions in a plane. By excluding collocation of the agents,
we define the state space of the system as
! ˇ )
Mstate “ g1 , g2 P SEp2q ˆ SEp2qˇg1 e3 ‰ g2 e3 (14)
ˇ

„ 
T xi yi ri
where e3 “ r0 0 1s and gi “ .
0 0 1
In terms of this Lie-group formulation the system dynamics can be represented as
` ˘
g9 i “ gi ξi pui q “ gi νi A1 ui ` A2 (15)

where A1 and A2 represent infinitesimal generators of planar rotation and translation, re-
spectively. Moreover A1 and A2 can generate the Lie-algebra under bracketing. As we are

5
interested in steering laws which leave our system dynamics invariant under rigid motion, we
can formulate a reduction to the shape space, a quotient manifold Mstate {SEp2q of relative
positions and velocities of the agents. We define g P SEp2q as
» fi
x1 ¨ x2 x1 ¨ y2 x1 ¨ r
g “ g1´1 g2 “ – y1 ¨ x2 y1 ¨ y2 y1 ¨ r fl , (16)
0 0 1

where r fi r2 ´ r1 denotes the baseline vector. Now the shape space for the planar two-agent
system can be defined as
! ˇ )
ˇ 2 2
Mshape “ g P SEp2qˇg13 ` g23 ‰0 . (17)

Moreover, g assumes a left-invariant dynamics on SEp2q as

g9 “ gξ (18)

where ξ “ ξ2 pu2 q ´ g ´1ξ1 pu1qg P sep2q, and the proposed control law (12) depends only on
the shape variable g as
„ 
g
ui “ µ īi , i P t1, 2u, ī “ t1, 2uztiu. (19)
νi
Therefore from [17] it can be concluded that the reduced dynamics (18) evolves on the shape
space Mshape .
y2
x2 x1
φ
y1
θ2
r2 r “ r2 ´ r1 ψ
ϑ θ1
r1
Figure 2: Illustration of scalar shape variables (ρ, ψ, φ) used to parametrize the shape space
Mshape .

4.1.2 Shape dynamics

Now, through polar parametrization, we introduce some geometrically meaningful scalar
variables to parametrize the shape space. By defining

r “ r2 ´ r1 “ ρeiϑ
x1 “ eiθ1 (20)
x2 “ eiθ2 ,

we introduce ψp“ π ´ ϑ ` θ1 q and φp“ θ1 ´ θ2 q to represent the relative orientation of the

velocity vectors. Clearly, ψ represents the relative orientation of x1 with respect to the

6
baseline vector r, and φ represents the misalignment in velocity directions. From (16) one
can notice that g P Mshape can be represented in terms of these scalar shape variables as
» fi
cos φ sin φ ´ρ cos ψ
g “ – ´ sin φ cos φ ρ sin ψ fl . (21)
0 0 1

Then, by a straightforward calculation, one can show that the shape dynamics are given in
terms of the scalar variables by

ρ9 “ ν1 cos ψ ´ ν2 cospψ ´ φq
1“
ψ9 “ ν1 u1 ´ ν1 sin ψ ´ ν2 sinpψ ´ φq
‰
(22)
ρ
φ9 “ ν1 u1 ´ ν2 u2 .

4.1.3 Analysis of topological velocity alignment

Here we consider a particular context of the two-agent planar system wherein each agent
employs the strategy for topological velocity alignment (TVA). In terms of the original state
variables the contrast functions take the form
1
Θi “ px ´ xi q ¨ pxī ´ xi q , i P t1, 2u (23)
2 ī
and the i-th agent is declared to attain TVA if Θi “ 0. From (23) one can immediately
notice equality of the contrast functions for both agents, and hence we define a common
contrast function Θ “ Θ1 “ Θ2 . Clearly, this common contrast function can be represented
in terms of scalar shape variables as

Θ “ 1 ´ cos φ, (24)

and hence we can conclude that Θ “ 0 if and only if φ “ 0.

Next, for this two-agent planar system we define a 2-dimensional TVA manifold MT V A Ă
Mshape as ! ˇ )
MT V A “ ρ, ψ, φ P Mshape ˇφ “ 0 . (25)
ˇ

Moreover, from (19) one can notice that the feedback law can be expressed in terms of shape
variables, taking the form ´ ¯
u1 “ ´ νµ1 sin φ
´ ¯
µ
(26)
u2 “ ν2
sin φ.
Clearly, the steering control becomes identically zero on the TVA manifold, and as a conse-
quence the mismatch in velocity direction remains identically zero (22). Now we will formally
introduce the notion of invariance.

Definition 4.2 (Invariant Manifold). A manifold M is said to be invariant under the flow
of a vector field X on M if for any m P M, Ft pmq P M for small t ą 0, where Ft p¨q denotes
the flow of X. One can show that this condition is equivalent to X being tangent to the
manifold M.

7
 If both agents employ a steering control of the form (26), the closed loop dynamics for a
two-agent planar system can be represented as

ρ9 “ ν1 cos ψ ´ ν2 cospψ ´ φq
1“
ψ9 “ ´µ sin φ ´ ν1 sin ψ ´ ν2 sinpψ ´ φq
‰
(27)
ρ
φ9 “ ´2µ sin φ.

We should note that prohibition on collocation, i.e. ρ ą 0, is not enforced by these dynamics.
Now we state the following proposition associated with the closed loop shape behavior.

Proposition 4.3. The topological velocity alignment manifold MT V A Ă Mshape is invariant

under the closed loop shape dynamics (27). Moreover if γptq P Mshape is a trajectory of (27)
which does not have a finite escape time, and Θp0q ‰ 2, then

Θptq Ñ 0 as t Ñ 8, (28)

i.e. γptq converges asymptotically to MT V A .

Proof. From (24) and (27) we have

9 “ φ9 sin φ “ ´2µ sin2 φ “ ´2µΘ 2 ´ Θ

` ˘
Θ (29)

and hence MT V A Ă Mshape is an invariant manifold under the closed loop shape dynamics.
This equation (29) also implies that the agents will keep on moving in the opposite direction
(Θptq “ 2) if their initial directions were opposite to each other (Θp0q “ 2).
Moreover, by assuming Θp0q P p0, 2q, we have

2e´4µt
Θptq “
C ` e´4µt
2
where the constant C is defined as C “ Θp0q
´1. Since e´4µt Ñ 0 as t Ñ 8, we have Θptq Ñ 0
as t Ñ 8.
Next we focus on the restricted dynamics on MT V A , and by substituting φ “ 0, (27)
yields ` ˘
ρ9 “ ν1 ´ ν2 cos ψ
1` (30)
ψ9 “ ´ ν1 ´ ν2 sin ψ.
˘
ρ
Now, by assuming ν1 ´ ν2 to be non-zero, the evolution of a phase plane trajectory can be
described as
dρ ρ cos ψ
“´ , (31)
dψ sin ψ
which can then be integrated to show that the quantity ρptq sin ψptq is conserved along any
trajectory on the TVA manifold (MT V A ). Hence it is clear that the region {ρ ą 0, π ą ψ ą 0}
(or similarly {ρ ą 0, 0 ą ψ ą ´π}) is positively invariant under the restricted dynamics (30).
The corresponding phase-portrait is shown in Fig 3.

8
 3

ψ 0

−1

−2

−3
0 2 4 ρ 6 8 10

Figure 3: Phase portraits for the restricted dynamics (30).

4.2 Algorithm for an n-agent system in a three-dimensional set-

ting
In this sub-section we focus toward TVA in its true sense. By bringing in an additional
neighbor into consideration whenever vCOM becomes zero, we propose an algorithmic way
(Algorithm 1) to implement TVA in a real system. Clearly, this provides a way to avoid
ill-posedness associated with vCOM being zero because non-zero speeds of individual agents
ensure that considering an extra neighbor will make an otherwise zero vCOM non-zero. More-
over, this approach towards flocking can easily be restricted to a planar setting, and it has
been demonstrated in what follows.

5 Implementation Results
5.1 Experimental Setup
Our experimental test-bed is comprised of Pioneer 3 DX wheeled robots from Adept Mo-
bileRobots [26]. These compact differential-drive mobile robots are equipped with reversible
DC motors, high-resolution motion encoders and 19cm wheels, and the onboard computation
is done via 32-bit Renesas SH2-7144 RISC microprocessor, including the P3-SH microcon-
troller with ARCOS. The sensors on the robot include eight forward-facing ultrasonic (sonar)
sensors. ARIA [27] provides an interface for controlling and receiving data from the robot,
and communication with the robot for sending control commands (forward velocity and turn-
ing rate) is done via 802.11-b/g/n networking. The width of the robot is 380 mm and it has
a swing radius of 260 mm.
Algorithm implementation (i.e, feedback law computation) has been done in C++ using
ROS [28], along with ROS-ARIA, as the interfacing robotics middleware. The experiments
have been carried out in a laboratory environment equipped with a sub-millimeter accurate
Vicon motion capture system [29]. We use a Dell workstation to run ROS, and this computer
is connected to the Vicon server via a dedicated Ethernet connection.
The Vicon system captures the motion of the robots and sends out the position and

9
 Algorithm 1: Topological Velocity Alignment - 3D
Data: Initial Time - tinitial ; Final Time - tf inal ; Sampling Interval - ∆; Number of
Agents - n; Initial Position and Orientation - tgi uni“1 ; Neighborhood Size - K
begin
Initialize: tcurrent ÐÝ tinitial ;
for i = 1 to n do
Initialize: State - Xi ÐÝ gi ;
while tcurrent ď tf inal do
for i = 1 to n do
Define: Ni - the set of K-nearest neighbors ;
Compute: Neighborhood center of mass velocity - vCOM ;
if vCOM “ 0 then
Define: Ni - the set of K ` 1-nearest neighbors ;
Compute: Neighborhood center of mass velocity - vCOM ;
Compute: Steering control - ui , vi ;
Implement: Steering Control - tui , vi uni“1 ;
Update: State - tXi uni“1 ;
Update: Time - tcurrent ÐÝ tcurrent ` ∆ ;

heading data to the computer running ROS. The control law program listens to this data,
and transmits the individual turning rates (as individual speeds are assumed to be constant).
Both of these operations are carried out at a frequency of 25 Hz. As the velocity vector (9rki ,
with k denoting the time index) is aligned along the robot heading, xki and yik can be directly
computed from the heading data. Then, the curvature variable uki is evaluated from the
corresponding control laws (8, 12), and the turning rate ωik pi “ 1, 2, . . . , nq (in degrees/sec)
is computed as: ˆ ˙
k 180
ωi “ νi uki . (32)
π
Next we will present the implementation results of the two control laws in our robotic
test-bed. In this paper, we are presenting results for which the speeds of all the individual
agents are same, i.e. νi “ νj , @i, j. Though it should be noted that both control laws can be
implemented with different speeds.

5.2 Implementation of MMC

Here we will describe the implementation results of MMC, difficulties of pure MMC law (5)
and usefulness of using the dissipative control term (8). First, the original MMC law (5) is
used to control the robots. The observed trajectories are then compared with theoretically
predicted ones. To generate the ideal trajectories in discrete time, we integrate the reduced
system dynamics (6). Considering the conserved quantity in the system (7), we used the
method described in [30] for integration instead of general ODE solver, which otherwise would
not be able to keep the quantity Epρ, γq constant to its initial value. From the integrated
values of ρk and γ k , we reconstruct the trajectories (i.e. ri k , xi k , yi k for i “ 1, 2) [13], where
k denotes the time index. At each time instance tk , the error (eki ) is then calculated as:

10
 1600 Robot1 1400 Robot1
Robot2 Robot2
1400
1200
1200
1000
1000
y (in mm)

y (in mm)
800 800

600
600
400
400
200
200
−1000 −500 0 500 −800 −600 −400 −200 0 200 400 600
x (in mm) x (in mm)

(a) Trajectory (Experiment) (b) Trajectory (Simulation)

250 0.4
Experimental
200 Simulation
0.38

Curvature control, u (in sec )

−1
150

100 0.36
γ (in mm/sec)

50
0.34
0

−50 0.32

−100
0.3
−150 Experimental
Simulation
−200
400 600 800 1000 1200 1400 1600 0 20 40 60 80 100 120
ρ (in mm) Time (in sec)

(c) Phase plot (d) Curvature

1200 5
Percent error in conserved quantity, E (%)

Robot1
Robot2 0
1000
−5

800 −10
Error (in mm)

−15
600
−20

400 −25

−30
200
−35

0 −40
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time (in sec) Time (in sec)

(e) Error (f) Percentage error in conserved quantity

Figure 4: Comparison between experimental and simulation results under pure MMC law in
action (with µ “ 0.001 mm´1 and ν1 “ ν2 “ 200 mm/sec)

eki “ |rki,expt ´ rki,ideal |.

The plots of a sample run using pure MMC law (5) are given in Fig 4. The parameters
for this run are, µ “ 0.001 mm´1 and ν1 “ ν2 “ 200 mm/sec. It can be seen from the figures

11
 600 600
Robot1 Robot1
400 Robot2 400 Robot2

200 200
0 0
y (in mm)

y (in mm)
−200 −200
−400 −400
−600 −600
−800 −800
−1000 −1000
−1200
−1200
−500 0 500 1000 −500 0 500 1000
x (in mm) x (in mm)

(a) Trajectory (Experiment) (b) Trajectory (Simulation)

300 −0.24
Experimental Experimental
Simulation −0.26 Simulation
200

Curvature control, u (in sec−1)

−0.28

100 −0.3
γ (in mm/sec)

−0.32
0
−0.34

−100 −0.36

−0.38
−200
−0.4

−300 −0.42
500 1000 1500 2000 0 100 200 300 400 500
ρ (in mm) Time (in sec)

(c) Phase plot (d) Curvature

250 8
Percent error in conserved quantity, E (%)

Robot1 Actual Error

Robot2 6 Cumulative Average
200
4
Error (in mm)

150 2

0
100 −2

−4
50
−6

0 −8
0 100 200 300 400 500 0 100 200 300 400 500
Time (in sec) Time (in sec)

(e) Error (f) Percentage error in conserved quantity

Figure 5: Performance of the system significantly improved upon addition of the dissipative
control term (with µ “ 0.001 mm´1 , ν1 “ ν2 “ 200 mm/sec and kd “ 1 ˆ 10´15 mm´6 sec3 )

that the experimental trajectories diverge from the ideal ones and thus the error also keeps
increasing over time in Fig 4e. This also affects the phase plot in Fig 4c as we can clearly see
the spiraling out type behavior as opposed to the ideal periodic behavior. Possible reasons

12
behind this have already been discussed in section 3.1.
To overcome the problem with original MMC law, a stabilization term has been intro-
duced in the control law (8). We applied the modified control law with kd “ 1 ˆ 10´15
mm´6 sec3 and kept all other parameters fixed. The resulting performance is quite satisfac-
tory as shown in Fig. 5a (refer [31] for implementation video). The error (Fig 5e) is also
bounded („ 250 mm) within the size of the robots („ 400 mm). Also, it can be seen from
Fig 5f that the cumulative average of the error in conserved quantity converges to almost
0%, indicating superior performance.

5.3 Implementation of TVA

We implemented the TVA control law (12) in a 2 dimensional setting (i.e. vi ptq is ignored).
As the implementation is in discrete time, we followed Algorithm 1 in our implementation
in order to avoid the singular case of |vCOM | “ 0. To demonstrate the performance of TVA
control law, we designed three different experiments (refer [31] for implementation videos).
In these experiments, the sonar sensors on the robots were activated to sense any obstacle
in the direction of motion of the robots and if any robot can sense such an obstacle, it will
simply apply a maximum turning rate (ω sat ) to avoid collision. The sonars are programmed
to detect an obstacle only in close proximity („ 300 mm) of the robots. In all our experiments
ω sat is taken to be 50 rad/sec, forward speeds of all of the robots are kept constant at 60
mm/sec and the value of the parameter µ is chosen to be 1 Hz.

5.3.1 Experiment 1
A system with eight agents is considered and we apply same TVA law to all of them.
The neighborhood size is taken to be three (i.e. K “ 3). The robots are initially placed
in arbitrary positions and řdirections. The footprints of the robots and the corresponding
contrast function, Θptq “ i Θi ptq is plotted against time in Fig 6a, 6b. The initial and
final directions of the robots are shown using arrows and the final positions of the robots
are denoted using dots. It can be seen from Fig 6b that the contrast function decays to zero
very quickly which indicates perfect velocity alignment within the swarm.

5.3.2 Experiment 2
Next we decreased the neighborhood size and made it one (K “ 1), so that every robot
only ‘communicates’ with its closest neighbor. We chose the initial positions in such a way
that they may form sub-clusters instead of moving as a single swarm. This behavior is
called ‘splitting’ in a swarm. From Fig 6c, we can clearly see that the swarm of eight robots
gradually split from each other and form three different clusters. It is to be noted that even if
all the agents are not going in the same direction, the contrast function still converges to zero
(Fig. 6d). This happens because each of the robots are aligned with their nearest neighbors
and hence each of the individual contrast functions (Θi ptq) are zero. This experiment may
explain the splitting phenomenon observable in nature.

5.3.3 Experiment 3
Lastly, we combined the above two experiments, and conducted an experiment using six
robots in a swarm and another robot as a predator. A separate computer was used for

13
 4.5

4
2000
3.5

1000 3
y (in mm)

2.5

Θptq
0
2

1.5
−1000
1

−2000 0.5

0
−3000 −2000 −1000 0 1000 2000 0 10 20 30 40 50
x (in mm) Time (in sec)

(a) Trajectories: 8 agents, flocking (b) Contrast function: 8 agents, flocking

1.4
1000
1.2
500
1
0
y (in mm)

−500 Θptq 0.8

−1000 0.6

−1500 0.4

−2000
0.2
−2500
0
−2000 −1000 0 1000 2000 0 5 10 15 20
x (in mm) Time (in sec)

(c) Trajectories: 8 agents, splitting (d) Contrast function: 8 agents, splitting

4 5

1000

Number of nearest neighbors

4
0
3
y (in mm)

−1000
Θptq

−2000 2

−3000
1

−4000
0 0
−5000 −4000 −3000 −2000 −1000 0 1000 2000 0 20 40 60 80
x (in mm) Time (in sec)

(e) Trajectories: 6 agents, perturbation (f) Contrast function: 8 agents, perturbation

Figure 6: Trajectory and contrast functions of TVA for (i) Experiment 1 (a),(b) with 8
agents, demonstrates flocking behavior (K “ 3); (ii) Experiment 2 (c),(d) with 8 agents,
describes the splitting behavior due to low neighborhood size (K “ 1), and (iii) Experiment
3 (e),(f) with 6 agents, shows perturbation can cause a swarm to split, the trajectory of
the perturbing agent is not shown. (µ “ 1 Hz and νi “ 60 mm/sec is kept fixed for all
experiments)

14
manual control of the ‘predator’ robot.
At the beginning, neighborhood size is kept at K “ 3, such that the ‘communication’
graph among the robots stays connected and they move as an entire swarm in a common
direction. When the swarm comes close to the predator, the neighborhood size is decreased
to one. As we are not using any onboard visual sensing and the sonar sensing is done only
in very close region („ 300 mm), the change in neighborhood size is made manually. From
Fig 6f, we can see that the change in neighborhood size takes place at around 20 seconds
and we can also see a tiny jump in the contrast function at that time. The predator then
slowly approaches to one of the agents in the swarm, which abiding to its collision avoidance
rule, turns to avoid the predator. In Fig 6e, the trajectories of the agents are drawn in
dashed lines before the occurrence of this event and in solid lines afterwards. The trajectory
of the predator robot in not shown in the figure. After creating the initial perturbation,
the predator is slowly moved through the swarm causing some subsequent disturbances.
These perturbations create a noticeable impact in the swarm. As the attacked agent turns,
its neighbor also tries to align itself with that agent and so does its neighbor. This goes
on until the communication graph becomes disconnected and a split in the swarm is then
observed [31] just like in Experiment 2. As we can see in Fig 6e, the swarm is divided in two
groups after the attack of the predator. The jumps in plot of the contrast function in Fig
6f symbolize the perturbations caused by the external agent. The contrast function value
eventually converges to zero after the members are aligned with their neighbors within each
subgroup.

6 Conclusion and Future Work

This paper has introduced a control strategy (TVA) which, based on local information, at-
tempts to align the individual velocities, and the theoretical analysis for a special case of
planar two-agent system has been complemented by experiments in a laboratory environ-
ment. Additionally, we chose MMC for implementation due to its dynamic coverage property
with potential in many domains (resource harvesting, environmental monitoring, search and
rescue missions etc.). Moreover, the exploitive nature of MMC (through coverage in a certain
annular region) augments the exploratory nature of TVA (through making the agents move
in a common direction). Future works will primarily focus on two areas, namely effect of a
beacon (influencing both agents) into MMC, and understanding the effect of a covert leader
(equipped with extra information) in a flock. Also, this study does not address how the
system behavior will change in a noisy (or uncertain) environment. In future, the authors
have plans to extend current results to study the effect of sensor noise (along the lines of
[32]) and perform robustness analysis.

Acknowledgments
The authors would like to thank P. S. Krishnaprasad and K. S. Galloway for their valuable
comments and feedback.

15
References
[1] U. Halder and B. Dey, “Biomimetic Algorithms for Coordinated Motion: Theory and
Implementation,” accepted for International Conference on Robotics and Automation
(ICRA 2015).

[2] R. M. Olberg, A. H. Worthington, and K. R. Venator, “Prey pursuit and interception in

dragonflies,” Journal of Comparative Physiology A, vol. 186, no. 2, pp. 155–162, 2000.

[3] A. Mizutani, J. S. Chahl, and M. V. Srinivasan, “Insect behaviour: Motion camouflage

in dragonflies,” Nature, vol. 423, no. 6940, pp. 604–604, 2003.

[4] K. Ghose, T. K. Horiuchi, P. S. Krishnaprasad, and C. F. Moss, “Echolocating bats

use a nearly time-optimal strategy to intercept prey,” PLoS Biology, vol. 4, no. 5, pp.
865–873, May 2006.

[5] C. Chiu, P. V. Reddy, W. Xian, P. S. Krishnaprasad, and C. F. Moss, “Effects of

competitive prey capture on flight behavior and sonar beam pattern in paired big brown
bats, Eptesicus Fuscus.” Journal of Experimental Biology, vol. 213, no. 19, pp. 3348–
3356, 2010.

[6] M. Nagy, Z. Akos, D. Biro, and T. Vicsek, “Hierarchical group dynamics in pigeon
flocks,” Nature, vol. 464, pp. 890–893, April 2010.

[7] M. Ballerini, N. Cabibbo, R. Candelier, A. Cavagna, E. Cisbani, I. Giardina, V. Lecomte,

A. Orlandi, G. Parisi, A. Procaccini, M. Viale, and V. Zdravkovic, “Interaction ruling
animal collective behavior depends on topological rather than metric distance: Evidence
from a field study,” Proceedings of the National Academy of Sciences, vol. 105, no. 4,
pp. 1232–1237, 2008.

[8] A. Cavagna, A. Cimarelli, I. Giardina, G. Parisi, R. Santagati, F. Stefanini, and

M. Viale, “Scale-free correlations in starling flocks,” Proceedings of the National
Academy of Sciences, vol. 107, no. 26, pp. 11 865–11 870, 2010.

[9] Y. Inada and K. Kawachi, “Order and flexibility in the motion of fish schools,” Journal
of Theoretical Biology, vol. 214, no. 3, pp. 371 – 387, 2002.

[10] E. W. Justh and P. S. Krishnaprasad, “Steering laws for motion camouflage,” Proceed-
ings of Royal Society A, vol. 462, no. 2076, pp. 3629–3643, 2006.

[11] P. V. Reddy, E. Justh, and P. S. Krishnaprasad, “Motion camouflage in three dimen-

sions,” in Proceedings of 45th IEEE Conference on Decision and Control (CDC), 2006,
pp. 3327–3332.

[12] M. Mischiati and P. Krishnaprasad, “Motion camouflage for coverage,” in Proceedings

of American Control Conference (ACC), 2010. IEEE, 2010, pp. 6429–6435.

[13] ——, “The dynamics of mutual motion camouflage,” Systems & Control Letters, vol. 61,
no. 9, pp. 894–903, 2012.

16
[14] J. Marshall, M. Broucke, and B. Francis, “Formations of vehicles in cyclic pursuit,”
IEEE Transactions on Automatic Control, vol. 49, no. 11, pp. 1963–1974, November
2004.

[15] K. S. Galloway, E. W. Justh, and P. S. Krishnaprasad, “Symmetry and reduction in

collectives: cyclic pursuit strategies,” Proceedings of the Royal Society A: Mathematical,
Physical and Engineering Science, vol. 469, no. 2158, 2013.
[16] T. Vicsek, A. Czirók, E. Ben-Jacob, I. Cohen, and O. Shochet, “Novel type of phase
transition in a system of self-driven particles,” Phys. Rev. Lett., vol. 75, pp. 1226–1229,
Aug 1995.

[17] E. W. Justh and P. S. Krishnaprasad, “Equilibria and steering laws for planar forma-
tions,” Systems & Control Letters, vol. 52, no. 1, pp. 25 – 38, 2004.
[18] A. Attanasi, A. Cavagna, L. D. Castello, I. Giardina, T. S. Grigera, A. Jelic, S. Melillo,
L. Parisi, O. Pohl, E. Shen, and M. Viale, “Information transfer and behavioural inertia
in starling flocks,” Nature Physics, vol. 10, no. 9, pp. 691–696, 2014.

[19] S. Thurrowgood, R. J. Moore, D. Soccol, M. Knight, and M. V. Srinivasan, “A bio-

logically inspired, vision-based guidance system for automatic landing of a fixed-wing
aircraft,” Journal of Field Robotics, vol. 31, no. 4, pp. 699–727, 2014.
[20] G. Vásárhelyi, C. Virágh, G. Somorjai, N. Tarcai, T. Szörényi, T. Nepusz, and T. Vicsek,
“Outdoor flocking and formation flight with autonomous aerial robots,” arXiv preprint
arXiv:1402.3588, 2014.
[21] P. S. Corbet, Dragonflies: Behavior and Ecology of Odonata. Comstock Publishing
Associates, 1999.
[22] M. Mischiati and P. S. Krishnaprasad, “Mutual motion camouflage in 3D,” in Proceed-
ings of the 18th IFAC World Congress, Milan, Italy, 2011, pp. 4483–4488.
[23] R. L. Bishop, “There is more than one way to frame a curve,” American Mathematical
Monthly, pp. 246–251, 1975.
[24] F. Fahimi, Autonomous Robots: Modeling, Path Planning, and Control. Springer, 2008.

[25] E. Justh and P. S. Krishnaprasad, “Steering laws and continuum models for planar
formations,” in Proceedings of 42nd IEEE Conference on Decision and Control (CDC),
Maui, Hawaii, 2003, pp. 3609–3615.
[26] Pioneer P3-DX. [Online]. Available: http://www.mobilerobots.com/researchrobots/pioneerp3dx.aspx

[27] ROS-ARIA. [Online]. Available: http://wiki.ros.org/ROSARIA

[28] Robot Operating System (ROS). [Online]. Available: http://www.ros.org

[29] Vicon motion capture system. [Online]. Available: http://www.vicon.com

[30] M. A. Austin, P. S. Krishnaprasad, and L.-S. Wang, “Almost poisson integration of
rigid body systems,” Journal of Computational Physics, vol. 107, no. 1, pp. 105–117,
1993.

17
[31] Implementation videos. [Online]. Available: http://ter.ps/ICRA2015

[32] K. S. Galloway, E. W. Justh, and P. S. Krishnaprasad, “Motion camouflage in a stochas-

tic setting,” in Proceedings of 46th IEEE Conference on Decision and Control (CDC),
Dec 2007, pp. 1652–1659.

18