Morphogenetic Multi-Robot Pattern Formation Using Hierarchical Gene Regulatory Networks
1. FOCAS workshop, 2nd September 2013, Taormina, Italy
Morphogenetic Multi-Robot Pattern Formation
Using Hierarchical Gene Regulatory Networks
Professor Yaochu Jin and Dr. Hyondong Oh*
Nature Inspired Computing and Engineering (NICE) Group
Department of Computing, University of Surrey, UK
*EC FP7 project: Genetically-programmable self-patterning swarm-organs (Swarm-Organ)
3. Introduction
• Multi-robot systems (MRSs) are to collectively accomplish complex tasks
that are beyond the capability of any single robot
in the presence of uncertainties or with incomplete information
where a distributed control or asynchronous computation is required
flexible, robust, and adaptive
Search and rescue, cooperative transportation, mapping, and monitoring
• Morphogenetic robotics is a new emerging field of robotics for selforganisation of swarm or modular robots
which employs genetic and cellular mechanisms, inspired from
Biological morphogenesis and gene regulatory networks (GRNs)
• Morphogenetic pattern formation which can be highly adaptable to
unknown environmental changes
5. Biological Morphogenesis
• Morphogenesis is a biological process in which cells divide and differentiate, and finally
resulting in the mature morphology of a biological organism.
• Morphogenesis is under the governance of a developmental gene regulatory network
(GRN) and the influence of the environment represented as morphogen gradients.
• Morphogen gradients are either directly present in the environment of fertilised cell or
generated by a few cells known as organisers.
Frames from digital 4D movie of C. elegans
embryo development.
Movements of epidermal cells (green) and
neurons (red) during epidermal enclosure of
C. elegans
6. Gene Regulatory Networks (GRNs)
A gene regulatory network is a collection of DNA segments that interact with other
chemicals in its own cell or other cells, thereby governing the expression rate at which
the genes are transcribed into mRNA and proteins
Gene Regulatory Network
activator
activator
g1
Gene 1
Negative
repressor
feedback
g2
Gene 2
Positive
feedback
activator
g3
Gene 3
A gene regulatory network with three genes
Transcriptional regulatory network
controlling metabolism in E. coli bacteria
7. Multi-Cellular Interactions
Cell 1
Cell 2
The genes create GRNs
that exhibit complex
dynamic behavior to
control development
+
-
+
-
Gene codes for cell actions:
divide, die, communicate,
change cell-type
+
+
-
Cell-cell communication is
achieved by diffusive coupling
Gene
9. Cell-Robot Metaphor
Multi-Cellular System
Multi-Robot Systems
Concentration of gene G1
x-position
Concentration of gene G2
y-position
Concentration of gene P1
Internal state in x-coordinate
Concentration of gene P2
Internal state in y-coordinate
Cell-cell interactions through
TF diffusion
Robot-robot local interaction
Morphogen gradient
Target pattern to be formed
10. I. Adaptive Pattern Formation Using a
Hierarchical GRN
• Biological organisers imply a temporal
/ spatial hierarchy in gene expression
– For morphogenetic robotics, hierarchy
facilitates local adaptation
– Improvement of robustness and
evolvability
• Two-layer H-GRN structure for target
entrapping pattern formation
– Layer 1: pattern generation
– Layer 2: Robot guidance
• GRN model parameters are evolved
using a multi-objective evolutionary
algorithm
14. II. Adaptive Pattern Formation Using HGRN with Region-based Shape Control
• Predefined Simple Shape
– Desired region as a ring and obstacle avoidance
– Single moving target tracking
Movement (pos. & vel.) of a target is assumed
to be known or can be estimated
[unknown/known target velocity]
• Complex Entrapping Shape from Layer 1
– Stationary target with neighbourhood size adaptation
Adjusted by sensing (max) and bumper range (min)
– Tracking of multiple moving targets
15. III. Adaptive Pattern Formation Using
H-GRN with Evolving Network Motifs
• Evolving layer with network motifs
– Utilise basic building blocks for gene
regulation: positive, negative, OR, AND,
XOR, etc.
– Evolving GRN structures with evolutionary
optimisation to find the GRN model which
entraps multiple targets efficiently
17. Conclusions
• Morphogenetic approach to self-organised adaptive multi-robot pattern
formation using a hierarchical GRN (H-GRN)
• Highly adaptable to environmental changes resulting from unknown
target movements
• Applications: contaminant/hazardous material boundary monitoring or
isolation and transport/herding target objects to a goal position
18. Future Research Direction
• More biologically –inspired approaches to swarm robotics
• Realistic distributed system considering a swarm of robots’ sensing /
communication / computation capability
• Implementation with swarm robot testbed
– Kilobot: a low cost scalable robot designed for collective behaviours
19. Swarm Robot Testbed
Comparison of Small Collective Robot Systems
Robot
Cost
(GBP)
Scalable
operation
Sensing
Locomotion
/ speed
Body
size (cm)
Battery
(hours)
1. Alice
30*
none
distance
wheel
/ 4 cm/s
2
80
(10*)
charge,
power, program
distance,
ambient light
vibration
/ 1 cm/s
3
2
3.5-10
2. Kilobot**
1
3-24
3. Formica
4. Jasmine
wheel
15*
none
ambient light
3
1.5
Kilobot – commercially available & inexpensive
/ N/A
system for testing collaborative behaviour in a
distance, bearing,
wheel
90*
charge
3
1-2
/ N/A
very large (> 100)light color of robots
swarm
3
4
5
5. E-puck**
600
none
camera,
distance, bearing
wheel
/ 13 cm/s
7.5
6. R-One
150*
none
light, accel/gyro, IR
sensors, encoders
wheel
/ 30 cm/s
10
N/A
charge,
power, program
distance, bearing,
camera, bump
wheel
/ 50 cm/s
12.7
3
8. SwarmBot (EPFL)
N/A
none
distance, bearing,
accel/gyro, camera
treel
/ N/A
17
4-7
7
8
6
7. SwarmBot (MIT)
6
1-10
*part cost only / **commercially available
21. Swarm Robot Testbed
Kilobot Specifications
• Locomotion
– 2 vibration motors (255 power levels)
– 1 cm/s & 45 deg/s
• Communication & Sensing
– Infrared light transmitter/receiver
3 bytes up to 7 cm away
Distance by signal strength
– Ambient light sensor
• Controller
– Atmega 328 Microprocessor
– C language with WinAVR compiler
22. Swarm Robot Testbed
Kilobot Scalability
• Controller board
– Send a new program to all Kilobots at once
– Control the Kilobots (pausing or power on/off)
– One-meter diameter area
• Kilobot charger
– Charge ten Kilobots at one time
• Applications
– Foraging, leader following, transport, and etc.
– Need to be fairly simple due to limited capabilities
*References: http://www.k-team.com/mobile-robotics-products/kilobot
http://www.eecs.harvard.edu/ssr/projects/progSA/kilobot.html
M. Rubenstein et al., Kilobot: A Low Cost Scalable Robot System for Collective Behaviors, IEEE ICRA, USA, 2012
M. Rubenstein et al., Collective Transport of Complex Objects by Simple Robots: Theory and Experiments, AA-MAS, USA, 2013