Summer Internship 2004:

Modular Robot Unit – Characterisation, Design

and Realisation

Elmar Dittrich,
Department of Mechanical Engineering,
Technical University Ilmenau,
Germany.

Supervisor: Prof. Auke Jan Ijspeert

BIRG, Logic Systems Laboratory,
School of Computer and Communication Sciences,
Swiss Federal Institute of Technology Lausanne.

16th August 2004

Contents

Contents 2

List of Figures 5

List of Tables 7

1. Introduction 8

1.1. Motivation and Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.2. General Properties of Modular Robots . . . . . . . . . . . . . . . . . . . . 9

1.3. Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3.1. M-TRAN II: Self-Reconfigurable Modular Robot . . . . . . . . . . 11

1.3.2. Polybot: A Chain Reconfiguration Robot . . . . . . . . . . . . . . 13

1.3.3. Telecube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.3.4. CONRO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.3.5. The Crystalline Robot and The Molecule Robot . . . . . . . . . . 17

1.3.6. I-Cube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.3.7. Fracta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.3.8. Atron (Hydra-Project) . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.3.9. Titech . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

1.3.10. Fractal Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2
 Contents

2. Characterisation and Design 24

2.1. Design Steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1.1. First Draft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1.2. Actuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1.3. Casing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.1.4. Connection Mechanism . . . . . . . . . . . . . . . . . . . . . . . . 29

2.1.4.1. General Goals . . . . . . . . . . . . . . . . . . . . . . . . 29

2.1.4.2. Realised Design . . . . . . . . . . . . . . . . . . . . . . . 30

2.1.5. Batteries and Electronics . . . . . . . . . . . . . . . . . . . . . . . 32

3. Software 34

3.1. Actual Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.1.1. Control File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.1.2. JAVA-GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4. Experiments 36

4.1. Testing the Servo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1.1. Static Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1.2. Dynamic Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1.3. Long Run Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.2. Tests with different structures with up to 6 modules . . . . . . . . . . . . 37

4.2.1. Modules moving by itself . . . . . . . . . . . . . . . . . . . . . . . 38

4.2.2. Three modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.2.3. Four Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.2.4. Five Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.2.5. Six Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3
 Contents

5. Conclusion and Outlook 45

5.1. Proposals for the prototype 2 . . . . . . . . . . . . . . . . . . . . . . . . . 45

5.1.1. Actuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

5.1.2. Casing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

5.1.3. Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Bibliography 47

Bibliography 47

A. Tables 52

A.1. Calibration values of the current modules . . . . . . . . . . . . . . . . . . 52

A.2. List of RC-servo traders . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

A.3. Bill of materials per module . . . . . . . . . . . . . . . . . . . . . . . . . . 53

B. Drawings (CAD) 54

4
List of Figures

1.1. LEGO Mindstorm ® (a), Fischertechnik (b) and Topobo (c) . . . . . . . . 9

1.2. M-TRAN module (a) and connection mechanism (b) . . . . . . . . . . . . 12

1.3. Polybot G3 prototype (a) and G2 in spider configuration (b) . . . . . . . 13

1.4. Telecube fully contracted (a) and CAD drawing of the design (b) . . . . . 16

1.5. The CONRO element as CAD design (a) and real prototype (b) . . . . . 17

1.6. The Crystalline Robot (a) and The Molecule Robot (b) . . . . . . . . . . 18

1.7. I-Cubes module (left) and examples of link motion (right) . . . . . . . . . 19

1.8. Fracta modules (3D) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.9. Atron physical prototype (a) and CAD model of attachment mechanism
(b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.10. Fractal Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.1. First draft of the element consisting of body and u-profile . . . . . . . . . 25

2.2. Main parts of a standard RC-servo (a) and servo S-71 used in the module
(b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.3. Flat RC-servo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.4. Sandwich design of the casing: distance bolts (b) and milled PCB plates
(a) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.5. Final Design of BirgModRob including batteries and motor . . . . . . . . 30

2.6. Different off-the-shelf mechanisms: Nylatch (a), Southco ® (b) and Cam-
loc (c) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.7. Two modules connected in 45 ° orientation . . . . . . . . . . . . . . . . . 32

5
 List of Figures

2.8. Electronic connectors and wiring schematic . . . . . . . . . . . . . . . . . 33

2.9. Communication with off-board electronics . . . . . . . . . . . . . . . . . . 33

3.1. Java Motion Interface for Modular Robot: (a) drawing board, (b) result
board, (c) buttons to select the modules, (d) loop counter (see previous
section) and (e) time step (prev. section) . . . . . . . . . . . . . . . . . . 35

4.1. Setup shows two modules and the printed angular disc . . . . . . . . . . . 37

4.5. Modules moving by itself pulling the controller board . . . . . . . . . . . . 38

4.6. Three modules hopping around because of the push of module no. 3 and
trajectory file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.7. Four modules moving like a sidewinder and trajectory file . . . . . . . . . 39

4.8. RoboDog jumping through the laboratory and trajectory file . . . . . . . 40

4.9. ”Rolling Track” with six modules and trajectory file . . . . . . . . . . . . 40

4.2. Static Motor accuracy in horizontal plane (a) and vertical plane (b).
Phi equals target angle. Figure (b) shows bigger errors
due to gravity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.3. Dynamic motor accuracy in horizontal plane. Load = 1 module (a), 2

modules (b) and 3 modules (c). Unfilled area shows canceled experiments
due to exploding errors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.4. Dynamic motor accuracy in vertical plane. Load = 1 module (a), 2 mod-
ules (b) and 3 modules (c). Unfilled area shows canceled experiments due
to exploding errors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

6
List of Tables

1.1. Global Properties of the listed modular robots . . . . . . . . . . . . . . . 11

1.2. Mechanical Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.3. Electronics and Communication . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1. Specifications of BIRG-ModRob . . . . . . . . . . . . . . . . . . . . . . . . 24

2.2. Selection of RC servos arranged by torque . . . . . . . . . . . . . . . . . . 27

A.1. Calibration of current servos used in the modules . . . . . . . . . . . . . . 52

A.2. List of servo brands and traders . . . . . . . . . . . . . . . . . . . . . . . . 52

A.3. Bill of materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

7
1. Introduction

A modular robot can be defined as a robotic system consisting of a set of discrete

components (or LEGO-like building blocks). This special field of robotics holds many
interesting challenges, not only in terms of mechanical engineering, especially in the
field of mechatronics (MEMS1 , smart actuators) as well as in terms of control strategies
(distributed control, autonomous strategies). There is also a parallel to biological sys-
tems: Built from only 100 different cell types, biological systems represent thousands of
different organisms and functionalities.

1.1. Motivation and Goals

Although the existing modular robots are still far away from any real application there
are interesting scenarios in which the robot could operate. Imagine a building that has
been damaged by earthquake, containing a variety of obstructions and possibly not being
suitable for any particular standard robot. A self-reconfigurable modular robot could
adopt to the current environment. Assuming it has enough sensory input and some
algorithm that would allow to change the shape immediately, it could crawl through
narrow holes. In the inside of the building it could reshape into a legged robot and walk
over rubble, climb stairs or even reach surfaces much higher than itself.
With our first prototype at BIRG (Biology Inspired Robotics Group) we follow up
simpler approach. By using mostly off-the-shelf technology, combined with a low-end,
but robust design we make the construction cheap and easy to manufacture. These
constraints result in the possibility to manufacture a huge number of pieces. State of the
art devices are used for communication (Bluetooth) and control (Field Programmable
Gate Array). Therefore, the circuit boards are already in development by another
student in the lab but should be integrated in the existing prototype.
After more then one simulations, concerning self-organisation of locomotion with help
of evolutionary algorithms were done at the lab, it was time to design some physical
1
abbr.: Micro-Electrical-Mechanical-Systems

8
1.2 General Properties of Modular Robots 1 Introduction

hardware. One can study multiple kinds of locomotion with the existing prototype.
When the communication will be integrated, aspects of distributed control strategies
and communication technologies can also be examined. The advantages of modular
robotics compared to conventional robots are enormous: it increases the types of robots
you can simulate, like snake-robots, several legged-robots and any arbitrary structure.
This includes the field of collective robotics and inter-robot behavior as well.

1.2. General Properties of Modular Robots

When looking at all the projects concerning modular robotics, even more simulations
then real physical solutions exist, due to the problems of manufacture and control of
the simulated systems. The very first ideas came from Von Neumann’s self-reproducing
automata from the 1960’s as the origin of cellular automata models [32]. Nevertheless,
it is very difficult to construct these models as real hardware. The next category can
be described as statically configured with the help of an operator. This reminds on the
famous robotic-invention-systems like LEGO Mindstorms ® and Fischertechnik’s
ROBO Mobile Set. A very recent application comes from the MIT Media Lab and is
called Topobo [21]. The biggest effort on the design of modular robots is put in self-
reconfigurable structures, sometimes called metamorphic. This means that the modules
can change their position and thus their connections automatically without manual help.
Self-reconfiguration also opens the possibility of self-repair.

(a) (b) (c)

Figure 1.1.: LEGO Mindstorm ® (a), Fischertechnik (b) and Topobo (c)

9
1.2 General Properties of Modular Robots 1 Introduction

Homogeneity vs. Heterogeneity Homogeneity of the modules means that all the ele-
ments of the system are identical. In a robot made of homogeneous reconfigurable units,
however, each unit can be replaced by any other unit by changing the connection. This
means, that the position of the module in the robot defines its function. It could act as
a head, leg or spine depending on its location in the robot. Homogeneity also has an
advantage concerning self-repair, because it need not be known which kind of module
failed. At last, the production cost could be lowered due to mass production, simple
inspection and maintenance procedures.

If a modular structure consists of several, non-identical parts, it is called heterogeneous.

Here the function of the modules defines its position in the robot. Although for ”real”
modularity it is essential to have identical elements, it is very difficult to transform this
into physical elements. Because in homogeneous structures every module must have
its own CPU, power supply, sensors and communication interfaces, it will be physically
large.

Chain- vs. Lattice-Structure There is another classification in which modular robots

can be separated. Chain-like structures remind on biology inspired robots like snakes,
spiders, centipedes. In this configuration, the modules are arranged in sequence but can
also branch off if desired. Although the chain-like robots are able to move in different
gaits or even roll, forming a wheel, they are mostly controlled centralized according to
gait-control tables.

One can compare a lattice structure like a grid on a checkerboard, where every square
defines a position of an element. Locomotion is provided by reconfiguring its arrange-
ment, thus moving its center of mass of the whole structure. It can be described more
like a flow of liquid. Of course these modules have to be homogeneous, because they
only have to know its neighbors and theoretically there are no obstacles that can not be
overcome, assumed that there are enough modules available. Lattice configuration offers
one important advantage: Because positions are limited to the discrete grid space, they
do not have to monitor all their cells to avoid different parts smacking into one another.
But for todays applications the locomotion of such architecture is too slow and far away
from flowing like water.

Distributed Control The problem of decentralized control in distributed robots (even

every dynamic reconfigurable network, like swarm robotic systems and distributed sensor
networks) can be divided in two tasks: how the position in a network topology of each

10
1.3 Survey 1 Introduction

module can be defined using only its local connection sensors, and how these coupled
modules collaborate their local actions to accomplish global effects such as locomotion
and reconfiguration.

1.3. Survey

This section shows a recent survey about research in modular robotics. It focuses on real
physical applications. Table 1.1 – 1.3 show the main technical specifications of the most
important developments in this huge field. The most terms should be self-explanatory.
The robots that are listed here represent 3D structures. That means, they can move in
all three spatial dimensions. More about 2D structures like the Metamorphic Robot by
Pamecha [19] and Fractum by Murata [16, 29] in the listed citations.

homo- lattice/chain self- docking/dedocking control-strategies

geneous reconfigurable
M-TRAN II yes lattice/chain yes magnetic centralized,
force/SMA, springs distributed
(CPG+GA)
Polybot(G3) yes chain yes mechanical/SMA centralized (gait
tables)
CONRO yes chain yes mechanical/SMA master-slave,
decentralized
(”hormones”)
The yes lattice yes mechanical (DC centralized,
Crystalline motor) distributed
(”PacMan”)
The yes lattice yes mechanical (DC centralized
Molecule motor)
Telecube yes lattice yes magnetic force/SMA –
I-Cubes yes lattice yes mechanical/SMA centralized
Fracta yes lattice yes mechanical (DC distributed
motor)

Table 1.1.: Global Properties of the listed modular robots

1.3.1. M-TRAN II: Self-Reconfigurable Modular Robot

M-TRAN II is the second prototype of the Modular Transformer developed by the Dis-
tributed Systems Design Research Group around Haruhisa Kurokawa. At time it is the
most impressive modular robot concerning design, automatically connecting and discon-
necting and locomotion. It is composed of two semi-cylindrical parts and a link working

11
1.3 Survey 1 Introduction

as actuator with two DOF. Even though the two actuated axis are aligned in parallel
which only allows rotations in one plane, multiple connection surfaces compensate this
disadvantage. The connection mechanism works with a combination of permanent mag-
nets, nonlinear springs and SMA actuators. Magnetic connectors have advantages in ac-
curacy during approach, but the polarity of magnets leads to male/female design. Figure
1.2 shows a M-TRAN II module and a figure of the connection mechanism [15, 12, 36].

(a) (b)

Figure 1.2.: M-TRAN module (a) and connection mechanism (b)

M-TRAN II is controlled centralized. One can set together any configuration on the PC.
It is checked for connectivity and collision. A motion planner for lattice-like flow was
developed working with the help of a motion-rule database making the cluster follow a
desired trajectory. For ”real” locomotion a combination of a neural CPG2 and GA3 for
optimization were proposed. A number of locomotion patterns were derived, which were
also experimentally verified [35, 15].

2
abbr.: Central Pattern Generator
3
abbr.: Genetic Algorithm

12
1.3 Survey 1 Introduction

1.3.2. Polybot: A Chain Reconfiguration Robot

Concerning mechatronics, integration, miniaturisation and use of MEMS technology,

Polybot (mainly generation 3) can be seen as the best developed. As shown in ta-
ble 1.2 it has high-tech motors, plenty of sensors and fully hermaphroditic connectors.
Even the communication is established via CAN (Controller Area Network) bus a high-
performance bus communication for distributed control. However, all this high-tech has
one big disadvantage caused by these numerous devices. There are no batteries included
therefore it can not work untethered. Figure 1.3 shows the latest prototype and legged
configuration of the former generation. The connection mechanism works with a ”latch
in the hole” principal (90° increments in orientation possible) and SMA actuators for
locking. But with two connection surfaces per module only snake-like configurations are
possible (without help of passive elements, as seen in the humanoid-videos) [13, 34] .

(a) (b)

Figure 1.3.: Polybot G3 prototype (a) and G2 in spider configuration (b)

The control of Polybot is centralized, just like M-Tran. Their modules consult a gait-
control table, to get unique messages tailored for each module. For this challenge they
propose a description method (NML) for describing actual configurations and Phase
Automata as a kind of event-driven state machines. Polybot seems to be the most

13
 1.3 Survey

DOF ”ability to connectors actuators gears (max. size [mm] weight water- material
move itself” torque) [g] proof
M-TRAN II 2 (rot) no male/female DC-motor reduction 66x132x66 400 no Delrin ®
(190Ncm?)
Polybot(G3) 1 (rot) no hermaphro- Maxon- planetary+Har- 50X50X45 200 no –
ditic pancake monic Drive
(brush- (150 Ncm)
less)
CONRO 2 rot= yes male/female Servos incl. (40 Ncm) length=100 115 no Delrin ®
pitch+ (Futaba
yaw) S3102)
The Crystalline 2 (lin.) no male/female toy-mini- rack and pinion 178x51x51 341 no –
motor (21
Ncm)
The Molecule 4 (rot) on partner with male/female Servos worm gear for 102x140x102 1400 no ABS plas-

14
opposite sex connectors tic
Telecube 3 (lin) no male/female Maxon- lead screw 60x60x60 300 no Ertalyte ®
DC-gear (12N)
motor
I-Cubes 3 (rot) no male/female Cirrus CS worm gear length=80 370 no rapid pro-
21-BB Hi totyping
Perf. Servo plastic
(FDM)
Fracta 6(12) no hermaphro- DC motor timing belts, length=265 7000 no –
rot ditic worm gear,
Harmonic
Drive

Table 1.2.: Mechanical Properties

1 Introduction
1.3 Survey 1 Introduction

sensors computing communication power ext. power/ price

power con- batteries
sump-
tion
M-TRAN II – BasicStamp serial (4800 4W yes/3.8 V, 700 –
II baud,async.) mAh, Li-Ion
Polybot(G3) Hall(joint Motorola CANbus – yes/no –
angle), PowerPC
4 accel., 555+ 1 MB
force, 4 IR ext RAM
CONRO IR, CMOS Stamp II(e) IR-serial (9600 1.9 W yes/yes –
camera baud)
The Crys- IR, Hall Hitachi serial, IR – yes/3V Li-Ion 300 $
talline (exp./contract.) HD64F3644H)
The – microprocessor no intermolec- – yes/no 1000 $
Molecule ular, RS-485
with worksta-
tion
Telecube elctr. con- – IR 1-2 W yes/no –
tacts, IR
I-Cubes mechanical microprocessor no intermolec- – no/yes –
encoders, (H11) ular, RS-232
contact with PC
switch
Fracta 6 contact sen- – – 7W yes/no –
sors, rotary
encoders

Table 1.3.: Electronics and Communication

skilled modular robot in mechatronic aspects. But the problem of control in distributed
systems is inadequately solved by Polybot [33, 38, 37].

1.3.3. Telecube

A linear actuated lattice-like robot called Telecube was developed by the group around
Yim at PARC. Like Polybot it shows mechatronic design as its best. Custom made lead-
screws (with telescoping tube linear actuators), different low-friction and light weighted
plastics and a combination of magnetic- and SMA-actuated connectors affect this design
(see Figure 1.4) [28, 31].

1.3.4. CONRO

CONRO was developed by a group around Wei-Min Shen and Peter Will at the Uni-
versity of Southern California in Los Angeles. Every module has two standard RC
servomotors corresponding to pitch and yaw and it is self-sufficient and autonomous, be-
cause it has all computation hardware, batteries and sensors on-board. A single module

15
1.3 Survey 1 Introduction

(a) (b)

Figure 1.4.: Telecube fully contracted (a) and CAD drawing of the design (b)

is able to wiggle its body (limited ability to move). An equation for the ratio between
module-size, weight, capacity and power consumption is proposed and considered during
the design. This leads to a continuous operation time of 35 minutes and the power of
the servo to lift one other module [4]. The docking works fully autonomously. It is
divided into three steps: open loop phase for moving the IR-sensors into a line of sight,
the closed loop phase with the help of the IR receiver signal and the inverse kinematics
and at last the entrapment phase to push the pins in the sockets and latches, performing
high-frequency small ”shaking” movements to overcome friction [26]. It needs 3 minutes
with a successful ratio of 80 % to connect from random snake structure. A disadvantage
is that the connection mechanism between the elements is not genderless. For more
information concerning the hardware refer to [3, 5].

For continual rediscovery of network topology an Adaptive Communication (AC) proto-

col is proposed. Every possible combination of connections can be crypted by a variable
(all together 33 for a CONRO module). This information is provided by the IR sensors
in the connection plates. When every module probes all its connectors in every cycle
of the AC routine the topology is known in at most two cycles, independent whether it
is a acyclic (root-tree-model) or a cyclic graph. For locomotion or reconfiguration the
concept of hormones, which one finds in many biological organisms, comes in to play .
A hormone is a special message which carries either direct servo-values or higher-level

16
1.3 Survey 1 Introduction

(a) (b)

©USC Information Sciences Institute

Figure 1.5.: The CONRO element as CAD design (a) and real prototype (b)

sequences which can trigger the servos or latches to accomplish, for example, a detaching
action. This variable passes every module, but it will be received only by those according
to a rulebase. A rulebase can be seen as a gait control table (or a set of instructions for
reconfiguration), but used for every configuration. For example, it does not matter if a
snake consists of four or 40 modules or if a hexapod ”loses” legs becoming a quadruped.
Additionally, the communication is much faster than a centralized control and it has
the advantages of being fault-tolerant, asynchronous, scalable.The questions remains
how to develop the appropriate rulebase for a particular global behavior automatically
[24, 25, 27].

It should be remarked that according to the homepage of the lab4 a new generation of
self-reconfigurable modules named SERE is planned.

1.3.5. The Crystalline Robot and The Molecule Robot

Both robots are the result of projects in self-reconfiguring robots conducted by the team
of Daniela Rus at Dartmouth Robotics Lab.

The Crystalline Robot is a two-dimensional self-reconfigurable modular robot system

composed of Atoms. The Atoms are square and actuate by expanding and contracting
by a factor of two in each dimension. It is together with Telecube, the only modular
4
http://www.isi.edu/robots/sere/

17
1.3 Survey 1 Introduction

robot in this survey with linear actuation. This makes it suitable for walking through
volume (not on the surfaces of a cube of modules, rather tunneling it). Respectively two
of the four faces contain passive (channels) or active (bars) connectors that lock by a
quarter-turn of the bar [2, 22, 23].

Describing the Molecule Robot is not uncomplicated, especially finding the actuated
parts and getting an idea about possible movements. Like everywhere in our universe
a molecule consists of atoms, in our case two, which are connected with a rectangular
bond. Every atom is able to rotate around this bond 180 degrees. Hereby, the right-
ankled bond leads to movements in 3D. Additionally, the only axis of the atom which is
perpendicular to the bond-connecting face and lies not in the same plane like this, can
rotate also 180 degrees. The connectors used for the Molecule Robot can be compared to
the gripper mechanism of Swarm-Bot5 . A non-backdrivable worm gear mechanism drives
gripper arms (male) to connect to certain corresponding points (female) on the surface.
Molecules either have all male components or all female components as connectors.
This design allows elements to move lattice-like only on a substrate with a contrary sex
then itself. Figure 1.6 shows the last prototype of The Crystalline Robot and a ”male”
Molecule.

rotating gripper (2x1 DOF)

rotating bond connection (2x1 DOF)

(a) (b)

Figure 1.6.: The Crystalline Robot (a) and The Molecule Robot (b)

To control The Crystalline Robot in a decentralized way the PacMan algorithm is pro-
posed. They took the concept of pellets (food in the PacMan video game) which code the
path each module has to follow in order to achieve global lattice-like motion. However,
5
Swarm-Bots: European Project (incl. EPFL/Lausanne) to study new approaches to the design and
implementation of self-organizing and self-assembling artifacts. See: http://www.swarm-bots.org/

18
1.3 Survey 1 Introduction

the completeness of the algorithm relies on a convex shape of the current configuration
which is difficult to maintain during locomotion [1].
Due to the L-shaped construction of The Molecule Robot, it is capable of convex and
concave 90-degree transition, which in practice allows the module to walk on a surface
and when it reaches a wall, to overcome the transition and cross the wall. A lot of
planning algorithms for configuration and trajectory planning have been applied to this
robot (incl. parallelization and scaffold algorithm). For more information refer to [11,
10].
Because of their lattice structure and behavior they can not yet show impressive move-
ments like chain robots, but they seem to be a very good platform for studying motion
planning, parallel computing and path optimization.

1.3.6. I-Cube

Ünsal and Khosla [30] have designed I-Cube a modular self-reconfigurable robotic system.
I-Cube is a partite system composed of a three DOFs link and a passive element as
connector (see Figure 1.7). In the right part of the figure you can see the attachment
capabilities: (a) move from one cube face to another, (b) move one cube while attached
to another and (c) move from one cube to another. The connection is similar to the
Crystalline Robot, instead of a bar a cross-shaped male/female mechanism is used.

Figure 1.7.: I-Cubes module (left) and examples of link motion (right)

This construction tends to discover new possibilities in design and actuation. The ap-
proach for a moving link is unique in this field. But one can not proof the current status

19
1.3 Survey 1 Introduction

of this project and it is the only one where no videos exist, showing real robot mo-
tion. More informations about control algorithms (incl. metamodule and multi-layered
planner) can be found in [20].

1.3.7. Fracta

Fracta is the 3D version on Fractum and made in 1998. It can be marked as the first
feasible 3D system with self-reconfigurability. In this system a unit has 6 rotational
DOF and 6 actuated connection mechanisms. These 12 DOFs in total are driven by one
DC motor embedded in the central body. Twelve electro-magnetic clutches are used to
connect the motor output axis to one of the DOF. A junglegym-like structure is built by
this unit, and it is reconfigurable by appropriate sequence of connection/disconnection
and rotation motion among the units (in or out of plane). The connection mechanism
is a gripper-like but genderless construction.

Figure 1.8.: Fracta modules (3D)

Because of the design (weight =7 kg and size) it was not built for impressive move-
ments. The more interesting was to study distributed algorithms, including description
methods of the current structure and path planning (stochastic relaxation and heuristic
algorithms).

20
1.3 Survey 1 Introduction

The drawback of this design with high DOFs is that each module is quite big and heavy.

1.3.8. Atron (Hydra-Project)

Atron was built by the Adaptronics Group, University of Southern Denmark, as part
of the HYDRA project funded by European’s Information Society Technologies (IST)
programme6 . Atron is a lattice-based system and it is built up by two half-spheres that
are able to rotate relative to each other around an axis that goes through the centre of
the module (see Figure 1.9). The elements have tangential connectors, four on every half-
sphere. Around the male/female mechanisms the casing is flattened for better alignment
during connecting. Changes in configuration are achieved by perpendicular alignment
of rotation axis’s of the elements performing 90 degree rotations around the equator.
Communication is based on infra-red devices integrated in the connectors.

(a) (b)

Figure 1.9.: Atron physical prototype (a) and CAD model of attachment mechanism (b)

From the controlling point of view there is only particular information. Some keywords
are: rule-based controllers, metamodules, virtual force-, gradient- and genetic algorithms
[6, 17, 18].

6
www.hydra-robot.com

21
1.3 Survey 1 Introduction

1.3.9. Titech

At the Tokyo Institute of Technology one can find a lot of research in modular robotics.
Not really self-reconfigurable, but modular, the famous ACM-snake by Hirose [14] was
developed at the same institute.

Another department created a pneumatic cellular robot, a lattice-like robot with bending
bellows as actuators mounted on every surface of the cube-shaped body. The bellow is
enclosed by a custom-made stabilization mechanism which allows 180 degrees rotation in
the two-dimensional plane. Also, they developed a selective valve mechanism to reduce
the number of mechanical elements. They showed automatic reconfiguration but the
system is notable only because of the interesting actuator [8].

Another cellular robot with linear actuation and sliding mechanisms is proposed by Inou
[9]. They implemented a two DOF slide motion mechanism, which one can imagine in
the real world as cubes sliding over their edges and showing different structures, even
under load.

1.3.10. Fractal Robots

The homepage of the Fractal Robots project7 shows ”impressive”, blurred animations
of their real robots. The elements are shiny aluminium blocks which seems to have no
connector or mechanism - but they slide magically over another. By the way, one can
buy this robot for 1000 $ per cube. For me it seems a little bit like a hoax, because no
papers about this project can be found and I discovered several topics in news groups,
having discussions about it. Figure 1.10 shows the cubes in different sizes.

7
http://www.ecu.pwp.blueyonder.co.uk/

22
1.3 Survey 1 Introduction

Figure 1.10.: Fractal Robots

23
2. Characterisation and Design

The general goal of this project, was to characterise, design and realise a modular robot
unit. To determine functional specifications we made a list of properties similar to tables
1.1 – 1.3. Table 2.1 shows the specifications for the first prototype.

Specifications BIRG-ModRob
homogeneous yes
lattice/chain chain
self-reconfigurable no
DOF 1–2
”ability to move itself” yes
computing power FPGA
communication Bluetooth
price per module 50–70 e

Table 2.1.: Specifications of BIRG-ModRob

Additional constraints are listed as follows:

ˆ using of as much off-the-shelf parts as possible

ˆ make the module itself modular, in the sense of being able to change the design by
replacing or adding parts

ˆ ”open-design” for electronics and sensors which still have to be implemented

ˆ time = 5 months for realisation of up to 10 prototypes

Especially the combination of the first and the last point in the list above made it a
challenge. due to the problem of finding parts from any area that fit the design and
could be delivered in time with a good price. The result of using off-the-shelf parts is
a so called open-design, which means no determination on custom made pieces, rather
changing the components easily.

24
2.1 Design Steps 2 Characterisation and Design

2.1. Design Steps

2.1.1. First Draft

Because of the multiple constraints, I decided for a pretty simple construction with one
degree of freedom (rotational) inspired by the shape of M-Tran II (see section 1.3.1)
and the mechanism of WormBot [7]. The primary motivation which points towards a
rotary drive is the chain-like behavior of the final structure. Added to this, that linear
movement can only be achieved by gears (rack-pinion and worm gear, lead-screw) or by
direct actuators (piezo, pneumatic-cylinder). Each of this possibilities can not satisfy
the given constraints, especially price and time. In Figure 2.1 you can see the very first
idea consisting of a main body with a round shaped side and an u-shaped profile for
transmission of the torque.

Figure 2.1.: First draft of the element consisting of body and u-profile

2.1.2. Actuator

Based on the result of the first design step we needed a rotary actuator with a good ratio
between torque, price, weight and size. I decided for an electromagnetic actuator because
working range, torque/weight-ratio and the availability in plenty of different shapes, sizes
and designs. From the huge field of electromagnetic drives I furthermore decided to use
standard RC-servos as known from airplane or car-models. I also considered DC-motors
in combination with a gear box which are in advance in terms of size and shape when
you are looking for some ”closed” integrated solution. Stepper motors have to be taken
in account too, but I did not find some small device in the given price-range.
RC-servos are used quite often in robotics. They have the advantage of being optimized
for different requirements concerning torque, size and weight. The most important fea-

25
2.1 Design Steps 2 Characterisation and Design

ture is the integrated position control consisting of potentiometer, controller (Mitsubishi

M51660L) and power circuits. Figure 2.2 shows the main parts. This leads to three in-
puts which are power supply (DC, 5–7 V), ground and a pulse-width modulated (PWM)
controller signal.

gearbox

potentiometer

DC motor

controller

(a) (b)

Figure 2.2.: Main parts of a standard RC-servo (a) and servo S-71 used in the module (b)

The PWM-signal has two important parameters: pulse length and frequency. The fre-
quency is either 20 Hz or 400 Hz and the pulse-length determines the position directly
via its zero-position value (mostly around 1.5 ms) and resolution (mostly 10 µs per 1 °).
From the work with different servos, even from the same brand it is known, that these
parameters are not the same and have to be calibrated for every servo. This does not
really support the homogeneity and the exchangeability in modular robotics. To over-
come this, parameters should be saved in any device directly according to each servo.
Another disadvantage is the box-like shape which is pretty standardized (recently one
can find so called ”flat-servos”; see Figure 2.3) and therefore affect the whole design the
most.

When you have to choose from the enormous amount of different servos and brands
there are two parameters which appear to be most important: maximum torque and
price. Due to the maximum torque is the weight because of metallic gears and bearings
(for higher torques) and the size (bigger motors). The price is almost linear to those
properties. To make a first calculation about the required torque you have to have first
estimations about size and behavior in later configurations of one element in the modular
robot. The latter means how far one element should be able to lift one other or more

26
2.1 Design Steps 2 Characterisation and Design

Figure 2.3.: Flat RC-servo

of the modules. The size is a very important question which can not be fully answered
yet in this early step of design. Therefore, a small calculation in has been done which
takes in account the measurements, weight and options to lift up to three other elements
(see Appendix). This result does not take into considerations moment of inertia and
acceleration for dynamical behavior but which means not a big error due to the size
of the modules and the speed. The results of the calculation show a range from 9.86
Ncm (lifting one module) to 73.8 Ncm (lifting three modules). Especially, the length
which means the distance from the center of rotation (motor shaft) to the center of mass
of the other element has the biggest influence on the required torque which should be
considered in following optimizations of this prototype. Table 2.2 shows a selection of
servos with needed torque range and price.

brand / type max. torque [Nm] price [e]

1. FMA Direct PS705MBB 62 Ncm 25.-
2. Simprop SES 640 2BB MG 64 Ncm 22.-
3. Servo S-71 J/R 73 Ncm 19.95
4. Kopropo PS 2174 FET 120 Ncm 70.-

Table 2.2.: Selection of RC servos arranged by torque

After my investigations including the European, American and Asian market I chose no.
3 from the table. Because of its torque, full metal gears and metal bearings (see Figure
2.2 (b)). A current special offer at Conrad Electronics, Germany for this type made it
really number one. The following list shows the main specifications:

ˆ max. torque: 73 Ncm

27
2.1 Design Steps 2 Characterisation and Design

ˆ max. speed: 60°/0.16 s

ˆ working range: 180 °

ˆ size: 41 x 20 x 41 mm

ˆ weight: 58 g

ˆ price: 19.95 e

I also tried to get information from different newsgroups, robotic boards and forums
in the Internet concerning reports on quality and behavior in real world of this servo.
But I did not find anything about the type I chose. In Section 2.2 you find more about
experiments with the servo under different conditions, including a long run test.

At the end of my investigations I found a servo with quite impressive properties (no. 2 in
Table 2.2). It has a similar torque like the one above, but it is smaller in size and weight.
We tested two of them and they seem to be quite good, despite of the inaccuracy under
upper range of load. At this point it makes no sense to have a much smaller solution,
because of open questions like batteries and electronic components which still have to
be integrated yet. But as an alternative for a second prototype this motor should be
further analysed.

2.1.3. Casing

As the most important device for the module has been made, the fixation of the motor
inside the module should be done. Because of the existing holes on the servo case made
for screws to fix it on substrate, a sandwich-design is proposed to achieve shape and
volume. As the main material for the case I chose printed circuit board PCB-material
(thickness = 2.4 mm), due to its material properties like stiffness, weight and available
manufacture technologies in workshops here at the EPFL. You can directly export your
data from your computer aided design (CAD)-tool to process them rapid-prototyping-
like into physical matter. I designed two plates, one with a cut for the servo and holes
for the fixation, hold in place by two hexagonal distance bolts made from Polyamid.
The shape of the plates and the height of the bolts determine surface and volume of
the element (see Figure 2.4). In this step 3D-CAD-tools, like ProEngineer ® show their
advantage in construction and design because one can change dimensions virtually but

28
2.1 Design Steps 2 Characterisation and Design

getting a clear impression about the influence of the overall design. We intended to have
scaled dimensions which means to have width, height and depth as a multiple of a what
we nearly realized (a = 45 mm and b = 2a, c = a).

c = 50

a = 45

(a)

(b)

b = 95 (incl. "U")

Figure 2.4.: Sandwich design of the casing: distance bolts (b) and milled PCB plates (a)

At last the box has to be closed to have possibilities for intermodular connections and
for protection. This is achieved by simply adding plates from the same material (thick-
ness = 1.6 mm) sidely directly screwed into the plastic distance bolts (with some custom
®
metal thread made by HELICOIL ) and at the other with the support of small alumi-
num profiles (see Figure 2.5).

To fix the aluminium u-profile properly, one side is screwed to normal plastic servo-
lever directly on the motor shaft, whereas on the other side an aluminium lever is fixed,
rotating the construction around an axle . For better sliding friction a bronze bushing
is pressed into the alu-lever. To keep the axle aligned with the servo shaft it is fixed in
a PCB plate on the head of the servo with the original screws of the servo-case.

2.1.4. Connection Mechanism

2.1.4.1. General Goals

The question of interconnecting robot elements can be seen as the key-question in mod-
ular robotics. A connection mechanism should therefore have the following properties:

ˆ power efficiency (no power consumption when connected)

29
2.1 Design Steps 2 Characterisation and Design

Figure 2.5.: Final Design of BirgModRob including batteries and motor

ˆ reliability (connections must endure various operations)

ˆ compact (mechanism must fit into a tight space)

ˆ genderless (allows to connect every connector among one other)

Most of the existing connection methods are based on penetration and shape matching,
where a latch is forced to slide into a groove or cut and is kept engaged by spring force.
For disconnecting one needs some active force generated mostly by shape memory alloy
(SMA) actuators (the only process which consumes power). The other option, you can
find it in the M-TRAN system, relies on connecting via magnetic force and disconnecting
with a help of a spring-SMA-actuator. The advantages of the mechanical solution are
highly reliable links which can withstand lateral forces mechanically, in return they need
very good alignment during the connection phase.

2.1.4.2. Realised Design

Additional to the general goals we wanted to have the possibility to interconnect the
elements in any particular angle mainly with a resolution of 45 degrees and on several
points on the surface of the module. Especially the last point is important for discovering

30
2.1 Design Steps 2 Characterisation and Design

structures useful for locomotion. The connector itself does not have to transmit any data
or energy because of wireless communication and autonomous battery power. On the
other hand, it is really difficult to find some off-the-shelf product as a connector which fits
into a tight place and is strong enough to withstand forces and torques in all directions
in space. We tried several fast locking mechanisms which you can see in Figure 2.6.
For connecting you have to insert the male part into the cut and do a quarter-turn.
The Nylatch ® works by pushing the pin into the hole which is bending up the nylon-
latches. But they do not fulfill the fact of being genderless. Imagine you mount the male
part of those connectors to the alu ”U” and the cam-locks inside the case - how do you
attache two back sides of the modules together? The solution would be mounting male
and female parts on every surface. However, you get into conflict with the goal having
different orientations.

(a) (b) (c)

® (b) and Camloc (c)

Figure 2.6.: Different off-the-shelf mechanisms: Nylatch (a), Southco

Finally, I decided for a combination of Velcro ® fastener, a very strong tissue which
works with principal of cockleburs and screws. The Velcro® can stand stand up against
sheer forces up to 600 N. Hence, you can fix up to three modules only with the Velcro® ,
especially when the surface is big enough like if you connect the sides of the modules.
Screws which support any complicated structure can easily be screwed into some metal
inlet with internal thread, which are pressed into the PCB plates. For first experiments
you can now achieve a reliable connection with an average mounting time of 45 sec per
connection.

Surely this is far away from having automatically connectible joints, but when first
experiments with this design show sufficient results and when all electronics are included,
it should be possible to create some custom made mechanism. Figure 4.1 shows two
modules, attached with the current connection mechanism.

31
2.1 Design Steps 2 Characterisation and Design

Figure 2.7.: Two modules connected in 45 ° orientation

2.1.5. Batteries and Electronics

With the question of batteries, which is an important issue for all autonomous robots, I
profited by the work which has been done for the salamander robot at BIRG. They use
rechargeable Li-Ion batteries with a nominal voltage of 3.7 V and capacity of 700 mAh.
With this capacity the module could theoretically work up to 45 min depending on the
load and the power consumption of the electronics. For the module we use two batteries
connected in serial to achieve an output voltage ranging from 6–8 V, which is slightly
higher then the input voltage of the servo, causing high velocity of the motor. They are
fixed laterally to the servo, aligned to the back of the element to shift the center of mass,
having less distance to the next motor shaft (when connected in serial).

For the first prototype I implemented a row of plugs fixed to the servo case. One wire is
used to disconnect the batteries from the servo, preventing loss of charge and to access
every battery individually for service and maintenance. The other plugs are used for
monitoring the battery voltage and to connect the off-board controller. Figure 2.8 shows
the open module with the batteries and a wiring schematic.

For the actual prototype the communication works as follows: any trajectory is created
on the PC (refer to next chapter) and transmitted via the serial interface to the Bluetooth
node. This device communicates with another BT-node which is connected to the FPGA,
which further on creates the PWM signal for the several servos. From the FPGA you
need two wires (signal and ground) leading to each module. In Figure 2.9 you can see
the schematic of this open loop control.

32
2.1 Design Steps 2 Characterisation and Design

battery 1 battery 2
− +
+ −

PWM

Figure 2.8.: Electronic connectors and wiring schematic

At the moment another student in the lab is working on the design and implementation
of all electronics (power-management, Bluetooth and FPGA). When this is finished the
modules should be completely autonomous (in the sense of power and communication)
and intermodulare distributed control via Bluetooth should be feasible.

BT−node
BT−node

Figure 2.9.: Communication with off-board electronics

33
3. Software

3.1. Actual Control

Because of the absence of sensors we have a straight forward open loop control. The
trajectories are created heuristically by trial and error (experiments show good results
even with very simple input). We implemented two possibilities to control the robot.

3.1.1. Control File

We propose a trajectory file with a self explaining syntax which can be seen as a gait
control table used in Polybot 1.3.2. Trajectory means a sequence of angle values over
time for each servo separately. We do not process in absolute time values, instead we use
fixed time steps (4t). This prevents the file from becoming too complex and we think
for our main purpose to study locomotion it should be more important to have looped
movements instead of a highly resolved trajectory. The following shows the syntax of
this file, named traj.txt:
/* Description of the movement or configuration */
<receiver> (addresses of the modules separated by space) </receiver>
<set delay> (time step in ms)
<set time factor> (factor multiplied with <set delay>)
<loop> (address of the loop) (loop counter) /*loop starts here*/

(motor values for each servo defined by <receiver> in the same order, separated by space)
</loop> (address of the loop) (value multiplied with time factor every loop cycle) /*loop ends
here*/

We have given addresses to the loops (the area between <loop> and </loop>) to make
it possible to interleave them. The time factor sets any value, which is multiplied with
the time step, to double or to halve the velocity of the movement. In Section 4.2 you
can see several examples for such a file.
This file can be edited by hand (as was done during all the experiments) or export the
®
angle values by any other program like Matlab , Xfig or MS-Excel.

34
3.1 Actual Control 3 Software

3.1.2. JAVA-GUI

For a more ergonomic way creating trajectories we wanted a graphical user interface
where the user is able to simply create any desired trajectory by clicking on the screen.
By chance a project at the lab had just been completed with the result of a GUI written
in Java. Originally this tool was thought as a plugin for Webots1 a platform for the
International Robot Judo Contest2 written by Jean-Philippe Egger. With the Java
Motion Tool it is possible to create trajectories and to save them. The program has
been modified to write the interpolated values into the above mentioned control file
and to send it to the robot achieving a corresponding movement. Additionally you can
connect the BT-nodes by pressing the ”Connect!” button. Figure 3.1 shows a screenshot
of this GUI. The board below is the drawing board and the upper board shows the
already drawn trajectories.

(e)

(d) (b)

(c)

(a)

Figure 3.1.: Java Motion Interface for Modular Robot: (a) drawing board, (b) result board, (c)
buttons to select the modules, (d) loop counter (see previous section) and (e) time
step (prev. section)

The number of modules in the select area can be easily increased.

1
a commercial 3D robot simulator by Cyberbotics Ltd. http://www.cyberbotics.com
2
see:http://www.roboka.org

35
4. Experiments

4.1. Testing the Servo

To verify the specifications of the motor, test the accuracy and long run behavior, a test
bench for the servo was constructed. Figure 4.1 shows the setup of the experiment.The
main goal was to get information about the accuracy under load. This was done with
the help of some printed semi-angular circle. The actual value was read of the scale.
Three different experiments were done to have sufficient information about the motor
in the future application. All experiments have been done under the same temperature
and light. The servo voltage was fixed to 7 V.

4.1.1. Static Test

Several angles were every time approached from the 0° position. Unfortunately, there
was no time driving every angle repeatedly to make statements about the statistic error.
We measured the actual angle and compared it to the desired angle under different load.
Additionally, the current was measured to keep the module in position. Figure 4.2 shows
the results.

Figure 4.2 shows a satisfying accuracy up to a load of three modules. But even a
maximum error of 7 degrees is not so much.

4.1.2. Dynamic Test

Here the servo was driving a sinus trajectory with adjustable amplitude and frequency.
The actual amplitude was read off the angular scale at the reversal points. This exper-
iment was done with a load of up to three modules, although with three modules only
small amplitudes with low frequency could be approached. Figure 4.3 – 4.4 show the
results. The non-coloured parts refer to canceled experiments to prevent the motor from
destruction. These areas can be seen as no-go area for the parameters.

36
4.2 Tests with different structures with up to 6 modules 4 Experiments

Figure 4.1.: Setup shows two modules and the printed angular disc

4.1.3. Long Run Test

A module was running for 7 hours without break, a load of two other modules doing
a sinus trajectory with an amplitude of 50° and a frequency of 0.6 Hz. We discovered
no abnormalities in movement and rhythm. A visual check up after the test showed
no changes. Also a short static test was carried out which showed no significant bigger
errors than before.

4.2. Tests with different structures with up to 6 modules

With the recent first prototype we were able to make the first tests of locomotion.
Inspired by existing configurations, famous in modular robotics, we made snake- and
inchworm-locomotion, several crawlers, polypeds and the rolling track. Surprisingly, we
had pretty impressive results, although we had to cope with the problem of asymmetric

37
4.2 Tests with different structures with up to 6 modules 4 Experiments

friction - which is a key question especially in crawling locomotion. Fortunately, the

shape (having multiple types of friction and coefficients) and some ”intelligent” speed
control (velocity dependant friction) could overcome this problem, so that we made the
most experiments on standard hard wood floor. Following we will show some configura-
tions and the control file. The figures showing screen-shots of videos we made and the
corresponding trajectory file.

4.2.1. Modules moving by itself

This experiment showed, that the modules are able to move by itself. This is a big
advantage and for a module with only one DOF, I have not found it during my research.
For this experiment a Styrofoam base was necessary to achieve asymmetric friction.

Figure 4.5.: Modules moving by itself pulling the controller board

4.2.2. Three modules

Picture 4.6 shows an interesting configuration with a passive pusher. This princip of
locomotion was the most successful configuration, even in the other experiments. This
indirect manipulation of inertia can help a lot overcoming the friction-problem.

38
4.2 Tests with different structures with up to 6 modules 4 Experiments

/*3-Heads2Side oft he middle*/

<receiver> 0 4 2 </receiver>
<set delay> 170
<set time factor> 1.0
<loop> 1 30
000
20 45 45
</loop> 1 1

Figure 4.6.: Three modules hopping around because of the push of module no. 3 and trajectory
file

4.2.3. Four Modules

As an example for a possible configuration you can see a sidewinder-like structure. This
name was given because it moves laterally.

/* 4-Head2Head and Heads2Tail */

<receiver> 3 2 4 0 </receiver>
<set delay> 100
<set time factor> 1.0
<loop> 1 35
-10 12 -12 10
-30 12 -12 30
-30 0 0 30
</loop> 1 1

Figure 4.7.: Four modules moving like a sidewinder and trajectory file

4.2.4. Five Modules

With five modules I tried a biomimetic way. Imagine a dog with immobile forelegs,
jumping because of flection in the spine and hind limbs. This configuration brought very
effective but pretty unstable locomotion.

39
4.2 Tests with different structures with up to 6 modules 4 Experiments

/* 5-Dog */
<receiver> 1 5 2 3 </receiver>
<set delay> 200
<set time factor> 1.0
<loop> 1 10
0 -45 -45 0
0000
</loop> 1 1

Figure 4.8.: RoboDog jumping through the laboratory and trajectory file

4.2.5. Six Modules

At last of the experiments we tried to make the rolling track, which has become pretty
famous in modular robotics. The basic principal is, to shift a pair of 90°-values around
the whole structure. Unfortunately we have only six modules, therefore the movement is
not very smooth and sometimes the robot changed the direction. It has be mentioned,
that especially with more modules (more DOFs), the heuristic method for creating the
trajectories becomes very hard.

/* 6-Rolling Track */
<receiver> 0 1 2 3 4 5 </receiver>
<set delay> 600
<set time factor> 1.0
<loop> 1 5
0 90 90 0 90 90
45 90 30 60 90 45
90 90 0 90 90 0
90 30 60 90 45 45
90 0 90 90 0 90
30 60 90 45 45 90
</loop> 1 1

Figure 4.9.: ”Rolling Track” with six modules and trajectory file

40
4.2 Tests with different structures with up to 6 modules 4 Experiments

All the here described experiments and much more interesting locomotion can be down-
loaded as video-files from birg.epfl.ch/page50163.html.

41
4.2 Tests with different structures with up to 6 modules 4 Experiments

Static behavior in horizontal plane

20
1 module
2 modules
15 3 modules

10

error [deg] 5

−5

−10

−15

−20
−80 −60 −40 −20 0 20 40 60 80
phi [deg]

(a)

Static behavior in vertical plane

20
1 module
2 modules
15 3 modules

10

5
error [deg]

−5

−10

−15

−20
−80 −60 −40 −20 0 20 40 60 80
phi [deg]

(b)

Figure 4.2.: Static Motor accuracy in horizontal plane (a) and vertical plane (b).
Phi equals target angle. Figure (b) shows bigger errors
due to gravity.

42
4.2 Tests with different structures with up to 6 modules 4 Experiments
Dynamic behavior horizontical [error in %]
90
55

80 50

45
70
40

60 35

phi [deg]
30
50
25

40
20

30 15

10
20
5

10
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
f [Hz]

(a)
Dynamic behavior horizontical [error in %]
90
35

80

30
70

25
60
phi [deg]

20
50

40 15

30 10

20 5

10
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
f [Hz]

(b)
Dynamic behavior horizontical [error in %]
90

35
80

70 30

60 25
phi [deg]

50 20

40 15

30 10

20 5

10 0
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
f [Hz]

(c)
Figure 4.3.: Dynamic motor accuracy in horizontal plane. Load = 1 module (a), 2 modules
(b) and 3 modules (c). Unfilled area shows canceled experiments due to exploding
errors.

43
4.2 Tests with different structures with up to 6 modules 4 Experiments
Dynamic behavior vertical [error in %]
90

60
80

70 50

60

phi [deg]
40

50
30

40

20
30

20 10

10
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
f [Hz]

(a)
Dynamic behavior vertical [error in %]
90
80

80
70

70
60

60
phi [deg]

50

50
40

40
30

30
20

20
10

10
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
f [Hz]

(b)
Dynamic behavior vertical [error in %]
90
20

80 18

70 16

14
60
phi [deg]

12
50
10

40
8

30
6

20 4

2
10
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
f [Hz]

(c)
Figure 4.4.: Dynamic motor accuracy in vertical plane. Load = 1 module (a), 2 modules (b) and
3 modules (c). Unfilled area shows canceled experiments due to exploding errors.

44
5. Conclusion and Outlook

This report describes the different design steps and considerations that lead to the cur-
rent design. Although some questions could not be fully answered the main goals of this
project were accomplished. An ”open” design is proposed which is roughly fixed in form
and function. The first experiments showed amazing results concerning the suitability
for modular robotics. As long as the key question of automatic connector mechanisms
remains, the robot can not be self-reconfigurable. Depending on the following appli-
cations the implementation of sensors should be considered. Most of these extensions
should be feasible with more or less expense on the current design.

5.1. Proposals for the prototype 2

5.1.1. Actuator

We are mainly satisfied with the used servomotor. The first experiments showed that
torque, speed and accuracy are sufficient. Especially the price makes it the best offer we
could get at this time. But because the market of standard RC-servos is changing rapidly
the offers should be checked frequently. In the appendix I put a list with the different
brands I found. Going along with optimizations in size and weight the Simprop-servo
(see: table 2.2) should be taken in consideration.

5.1.2. Casing

Depending on the integration of the Bluetooth, FPGA and the power circuit the op-
timization in size (especially the length) of the element should be taken in account.
Decreasing the length would bring a much less torque for lifting a module. To win some
space for sensors or additional circuits, the axle for the actuated ”u” should be directly
integrated into the upper layer of the casing. Another possibility would be to make the
”U” smaller in length and to integrate it into the casing to have real plain surfaces over

45
5.1 Proposals for the prototype 2 5 Conclusion and Outlook

the whole construction. However, it has to be mentioned that most of the locomotion
we showed is due to the ”active” support of this part.

5.1.3. Sensors

During the design process I was thinking a lot about information which one needs for
experiments with locomotion and distributed control. Most of the sensors you find
in modular robotics are due to the problem of self-reconfiguration (approaching the
connectors) and communication. We are not in charge of these problems yet and so I
was looking for other important data of the system which should be known.

Contributing to another project at the lab, which is about self-organisation in locomotion

with the help of numerical search methods, it would be great to make real simulations
with hardware in the loop. Therefore, you need information about the position of the
robot over time. Absolute position could only be achieved by tracking sensors and
orientation with gyroscopes or inclinometers.

Touch- or whisker-like sensors could be used to get information about collision between
elements or stance status in a possible configuration. I discovered a piezo-sensitive foil1
which could be attached to the corners or to the whole surface. Another possibility
would be using small electret microphones with the help of intelligent signal processing.

One sensor already integrated in the servo is the position potentiometer of the motor.
Attaching an additional wire can give either information about the actual position, or
take its derivative for knowing if a desired position is reached. An additional feature like
you find in Tobobo could be to manipulate the elements manually, to record the values
and to replay the trajectory.

Which of these possibilities should be implemented remains a future challenge for the
next prototype.

1
Images SI Inc.

46
Bibliography

[1] Z. Butler, S. Byrnes, and D. Rus. Distributed motion planning for modular robots
with unit-compressible modules. In IEEE International Conference on Intelligent
Robots and Systems, volume 2, pages 790–796, 2001.

[2] Z. Butler, R. Fitch, D. Rus, and Y. Wang. Distributed goal recognition algo-
rithms for modular robots. In Proceedings of the IEEE International Conference on
Robotics and Automation, volume 1, pages 110–116, 2002.

[3] A. Castano, A. Behar, and P. Will. The conro modules for reconfigurable robots.
IEEE Transactions on Mechatronics, 20:100–106, 2002.

[4] A. Castano, W.M. Shen, and P. Will. Conro: Towards deployable robots with
inter-robot metamorphic capabilities. Journal of Auonomous Robots, 24:879–888,
2000.

[5] A. Castano and P. Will. Mechanical design of a module for autonomous reconfig-
urable robots. In Proceedings of IEEE/RSJ Intl. Conf. of Intel. Robots Systems,
pages 2203–2209, 2000.

[6] D.J. Christensen, E.H. Østergaard, and H.H. Lund. Metamodule control for the
atron self-reconfigurable robotic system. In Proceedings of the 8th Conference on
Intelligent Autonomous Systems (IAS-8), pages 685–692, 2004.

[7] J. Conradt and P. Varshavskaya. Distributed central pattern generator control for
a serpentine robot, icann 2003. In ICANN 2003, 2003.

[8] N. Inou, H. Kobayashi, and M. Koseki. Development of pneumatic cellular robots

forming a mechanical structure. In Proceedings of the 7th International Conference
on Control, Automation, Robotics and Vision, ICARCV 2002, pages 63–68, 2002.

47
 Bibliography

[9] N. Inou, K. Minami, and M. Koseki. Group robots forming a mechanical structure-
development of slide motion mechanism and estimation of energy consumption of
the structural formation. In Proceedings of the 2003 IEEE International Symposium
on Computational Intelligence in Robotics and Automation, volume 2, pages 874–
879, 2003.

[10] K. Kotay and D. Rus. Algorithms for self-reconfiguring molecule motion planning.
In Proceedings of the IEEE International Conference on Intelligent Robots and Sys-
tems, volume 3, pages 2184–2193, 2000.

[11] K. Kotay, D. Rus, M. Vona, and G. McGray. Self-reconfiguring robotic molecule.

In Proceedings of the IEEE International Conference on Robotics and Automation,
volume 1, pages 424–431, 1998.

[12] H. Kurokawa, A. Kamimura, S. Murata, E. Yoshida, K. Tomita, and S. Kokaji.

M-tran ii: Metamorphosis from a four-legged walker to a caterpillar. In Proceedings
of the Conference on Intelligent Robots and Systems(IROS 2003), volume 3, pages
2454–2459, October 2003.

[13] Yim Mark, Duff David G., and Roufas D. Polybot: a modular reconfigurable
robot. In Proceedings - IEEE International Conference on Robotics and Automation,
volume 1, pages 514–520, 200.

[14] M. Mori and S. Hirose. Development of active cord mechanism acm-r3 with agile
3d mobility. In Proceedings of the 2001 IEEE/RSJ International Conference on
Intelligent Robots and Systems, volume 3, pages 1552–1557, 2001.

[15] S. Murata, A. Kurokawa, H.and Kamimura, E. Yoshida, K. Tomita, and S. Kokaji.

M-tran: Self-reconfigurable modular robotic system. IEEE Transactions on Mecha-
tronics, 7(4):431–441, December 2002.

[16] S. Murata, H. Kurokawa, and S. Kokaji. Self-assembling machine. In Proceedings

of the 1994 IEEE International Conference on Robotics and Automation, volume 1,
pages 441–448, 1994.

[17] E.H. Ostergaard and H.H. Lund. Evolving control for modular robotic units. In Pro-
ceedings of the 2003 IEEE International Symposium on Computational Intelligence
in Robotics and Automation, volume 2, pages 886–892, July 2003.

48
 Bibliography

[18] H.H. Ostergaard, E.H.and Lund. Distributed cluster walk for the atron self-
reconfigurable robot. In Proceedings of the 8th Conference on Intelligent Au-
tonomous Systems (IAS-8), pages 291–298, 2004.

[19] A. Pamecha, C.J. Chiang, D. Stein, and G. Chirikjian. Design and implementa-
tion of metamorhic robots. In Proceedings of the 1996 ASME Design Engineering
Technical Conference and Computers in Engineering Conference, pages 1–10, 1996.

[20] K.C. Prevas, C. Unsal, M.O. Efe, and P.K. Khosla. A hierarchical motion planning
strategy for a uniform self-reconfigurable modular robotic system. In Proceedings
of the IEEE International Conference on Robotics and Automation(ICRA ’02, vol-
ume 1, pages 787–792, May 2002.

[21] H.S. Raffle, A.J. Parkes, and H. Ishii. Topobo: A constructive assembly system with
kinetic memory. In Conference on Human Factors in Computing Systems, Vienna,
April 2004.

[22] D. Rus and M. Vona. Physical implementation of the self-reconfiguring crystalline

robot. In Proceedings - IEEE International Conference on Robotics and Automation,
volume 2, pages 1726–1733, 2000.

[23] D. Rus and M. Vona. Crystalline robots: Self-reconfiguration with compressible

unit modules. Autonomous Robots, 10(1):107–124, 2001.

[24] B. Salemi, W.M. Shen, and P. Will. Hormone controlled metamorphic robots. In
Proceedings of the IEEE Intl. Conf. on Robotics and Automation, pages 4194–4199,
2001.

[25] W.M. Shen, B. Salemi, and P. Will. Hormone-inspired adaptive communication

and distributed control for conro self-reconfigurable robots. IEEE Transactions on
Robotics and Automation, 18(5):1–12, October 2002.

[26] W.M. Shen and P. Will. Docking in self-reconfigurable robots. In Proceedings of the
2001 IEEE International Conference on Intelligent Robots and Systems, volume 2,
pages 1049–1054, 2001.

[27] K. Støy, W.M. Shen, and P. Will. A simple approach to the control of locomotion in
self-reconfigurable robots. Robotics and Autonomous Systems, Volume 44(3-4):191–
199, September 2003.

49
 Bibliography

[28] J.W. Suh, S.B. Homans, and M. Yim. Telecubes: Mechanical design of a mod-
ule for self-reconfigurable robotics. In Proceedings of the 2002 IEEE International
Conference on Robotics and Automation Washington, DC, pages 4095–4101, May
2002.

[29] K. Tomita, S. Murata, H. Kurokawa, E. Yoshida, and S. Kokaji. Self-assembly

and self-repair method for a distributed mechanical system. IEEE Transactions on
Robotics and Automation, 15(6):1035–1045, 1999.

[30] C. Unsal and P.K. Khosla. Mechatronic design of a modular self-reconfiguring

robotic system. In Proceedings of the IEEE International Conference on Robotics
and Automation(ICRA ’00), volume 2, pages 1742–1747, April 2000.

[31] S. Vassilvitskii, M. Yim, and J. Suh. A complete, local and parallel reconfiguration
algorithm for cube style modular robots. In Proceedings of IEEE International
Conference on Robotics and Automation(ICRA ’02), volume 1, pages 117–122, May
2002.

[32] J. von Neumann. Therory of Self-reproducing Automata. Univ. of Illinois Press,

1966.

[33] M. Yim, K. Roufas, D. Duff, Y. Zhang, C. Eldershaw, and S. Homans. Modular

reconfigurable robots in space applications. Autonomous Robots, 14(2–3):225–237,
2003.

[34] M. Yim, Y. Zhang, K. Roufas, D. Duff, and C. Eldershaw. Connecting and discon-
necting for chain self-reconfiguration with polybot. IEEE/ASME Transactions on
Mechatronics, 7(4):442–451, 2002.

[35] E. Yoshida and A.and Tomita K.and Kurokawa H.and Kokaji S. Murata,
S.and Kamimura. Evolutionary synthesis of dynamic motion and reconfiguration
process for a modular robot m-tran. In Proceedings of IEEE International Sympo-
sium on Computational Intelligence in Robotics and Automation, volume 2, pages
1004–1010, July 2003.

[36] E. Yoshida, S. Murata, A. Kamimura, and H.and Kokaji S. Tomita,

K.and Kurokawa. A self-reconfigurable modular robot: Reconfiguration planning
and experiments. The International Journal of Robotics Research, 21(10-11):903–
915, October–November 2002.

50
 Bibliography

[37] Y. Zhan, M. Yim, and D.and Roufas K. Eldershaw, C.and Duff. Scalable and recon-
figurable configurations and locomotion gaits for chain-type modular reconfigurable
robots. In Proceedings of the 2003 IEEE International Symposium on Computa-
tional Intelligence in Robotics and Automation,, volume 2, pages 893–899, 2003.

[38] Y. Zhang and M. Roufas, K.D.and Yim. Software architecture for modular self-
reconfigurable robots. In Proceedings of the 2001 IEEE/RSJ International Confer-
ence on Robotics and Automation, volume 4, pages 2355–2360, 2001.

51
A. Tables

A.1. Calibration values of the current modules

module - no. zero-position [ms] factor (for FPGA)

0 1480 90
1 1480 95
2 1550 96
3 1500 97
4 1580 97
5 1570 98

Table A.1.: Calibration of current servos used in the modules

A.2. List of RC-servo traders

company homepage
Höllein www.hoellein.com
Fliegerland Shop www.fliegerland-shop.de
Conrad Electronic GmbH www.conrad.de
Beck-Modellbau www.modellbau-beck.de
KO PROPO USA, Inc. www.kopropo.com
Modellbau Bichler www.modellbau-bichler.de
Otto Modellbau www.a1-modellbau.de
HOPE-Modellbau AG www.hopmodell.ch
MODELCO SA www.modelco.ch
FMA Direct www.fmadirect.com

Table A.2.: List of servo brands and traders

52
A.3 Bill of materials per module A Tables

A.3. Bill of materials per module

No. part quantity misc.

1 lower-layer 1 PCB, ProE: lower-layer.prt
2 upper-layer 1 PCB, ProE: upper-layer.prt
3 closure disc right 1 PCB, ProE: plaque-side-right.prt
4 closure disc left 1 PCB, ProE: plaque-side-left.prt
5 closure disc rear 1 PCB, ProE: plaque-tail.prt
6 servo plate 1 PCB; ProE: servo-plate.prt
7 fastener for rear disc 2 ALU, ProE: plaque-fix-tail.prt
8 U-profile 1 ALU, ProE: u-profile-1.prt
9 SERVO-S71 J/R 1 Conrad Electronic GmbH, order-no.: 224456
10 bronze bushing 1 Conrad Electronic GmbH, order-no.:237027
11 ALU-servo-lever 1 Conrad Electronic GmbH, order-no.:226636
12 Polyamid distance bolts 40 x M3 4 www.vogt.ch, order-no.:M30040.00.PA
13 axle 3mm 1
14 machine screws M2 x 6 mm 6
15 machine screws M2 x 8 mm 4
16 machine screws M2 x 4 mm 2 countersunk
17 machine screws M3 x 8 mm 8 countersunk
18 machine screw M3 x 6 mm 2 countersunk
19 machine screw M2.5 x 4 mm 4 countersunk
20 Velcro fastener
21 machine screw M3 x 8 mm 1 hexagon socket head

Table A.3.: Bill of materials

53
B. Drawings (CAD)

54