Using Internal States and Swarming Behaviours to Enhance the Performance
of Swarm Robots
Mohamed H. Mabrouk Colin R. McInnes
Department of Mechanical Engineering, University of Strathclyde, Glasgow.
{mohamed.mabrouk, colin.mcinnes}@strath.ac.uk
Great. No doubt this work has useful
Swarm robots?! What do you mean?
applications. Can you use it to solve mazes?
Swarm robotic systems are teams of
identical robots that have high rates of
information exchange among the a b c d
individuals. They have wide range of
Fig. 1 100 robot swarm during Fig. 2 Swarm of robots finds applications from seeking for survivors in
navigation (MIT, CSAIL) an area of interest (INL)
an earthquake site to planet exploration.
Right, so what is the problem? e f g h
Fig. 10 Maze application for a single agent using the internal state model dynamic potential (●), and that
with static potential (▲)
The work presented
To make path-planning for these enables to solve
systems, a computationally simple complicated path-
method is needed especially for planning problems such
increasing number of constituents. as Multi-level mazes for
The conventional artificial single robot, swarm of
potential field method is suitable. a b c d
robots using the swarm
However, this method has a fatal aggregation concept
Fig. 3 Deep local minimum forms behind a Fig. 4 Shallow local minimum forms
deep C-shape obstacle behind a shallow C-shape obstacle disadvantage which is the local and swarm of robots
minima formation. Previous that use the emergent-
attempts to solve the problem are like behaviour.
either computationally expensive
and then not suitable for real e f g h
Fig. 11 Maze application for agents using the internal state model; (*), and those with a static potential; (∆)
applications or works for easy
types of local minima.
Fig. 5 Behaviour of a conventional swarm of agents with fixed internal states
a b c
Swarming Behaviours
Oh! So
Leader-follower Aggregation Emergent behaviour what’s in
relationship behaviour
(solve the problem individually) (aggregation to face a problem)
(pattern formation) your mind?
Inspiring ideas from
swarming behaviour in d e f
natural systems to Fig. 12 Maze application for agents using the emergent-like model
enhance the performance
of the swarm robots. How clever. I am sure this work
Fig. 6 In nature we can see a flock of birds follow a leader, a colony of bees
aggregates to beat an intruder and a fish school provides a milling pattern
contributes to robotic field.
Performance comparison
Really! Have you succeed? tests prove that the method
is more efficient than the
most updated way to solve
Swarm leader: Swarm aggregation: Emergent-like behaviour: the local minimum problem
for robot navigation;
When one of the agents have When a Part of the swarm is Using the swarm center velocity to Fig. 13 Performance comparison test
Forward chaining.
clear way to the goal it has repelled the entire swarm increase the swarm members’
higher attraction potential aggregate to face the problem as a perception about the environment
then become temporarily flock to reduce the opportunity of enables the swarm to effectively
But wait a minute. I don’t believe that these ideas
leader to the other individuals leaving some agents stuck in the follow the obstacle wall boundaries can be used for real robots.
of the swarm. trap. by activating pure rolling motion of
the vortex pattern into which the Transparent C-shape
agents self-organize .
obstacle
Internal state agent
(Leader agent)
Trapped agent
Fig. 15 One of the used robots
G
G Of course it can. A ‘rescue’
a b c
mission application using the
swarm leader concept is
successfully implemented in a
swarming behaviour test-bed;
University of Strathclyde. The
test-bed has been built to test
the behaviour of identical
f
d e
robots, which interact via pair-
Fig. 14. Swarm leader concept implementation
wise interactions.
Conclusions please. For the files.
New potential field method has been presented which uses the concept of an agent internal state along
with the swarming behaviour to allow agents to manipulate the potential field in which they
manoeuvre. The method allows a swarm of agents to escape from and manoeuvre around a local
minimum in the potential field to reach a goal regardless of the formed local minima in the global
potential. Simulation results and implementations confirm the efficiency and applicability of the new
method.
Do you have publications in specialized
international conferences?
•Mabrouk M. and McInnes C. (2007). Swarm robot social potential fields with internal agent dynamics. In
Proceeding of the 12th International Conference on Aerospace Sciences & Aviation Technology (ASAT12), ROB-
02:1-14.
•Mabrouk M. and McInnes C. (2008a). Social potential model to simulate emergent behaviour for swarm robots.
Accepted by the 13th International Conference on Applied Mechanics and Mechanical Engineering, AMME-13.
Fig. 9 Emergent-like concept
•Mabrouk M. and McInnes C. (2008b). Wall following to escape local minima for swarms of agents using internal
Fig. 7 Swarm leader concept states and emergent behaviour. Accepted by the International Conference on Computational Intelligence and
Intelligent Systems, ICCIIS 2008. (Nominated for best paper prize, nominated to be included as a chapter of a book
Fig. 8 Swarm aggregation concept about state of the art in swarm robots).
