|
Control >> Aggregation
Aggregation
Aggregation is of particular interest since it stands as a
prerequisite for other forms of cooperation. For instance, in order to
assemble into a swarm-bot, s-bots should first be able
to aggregate.
Therefore, the aggregation ability can be considered as the
precondition for other tasks that the swarm-bot is expected to
be able
to carry out.
At the beginning of our research work, we have focused on the
study of aggregation mechanisms by exploiting a behavior-based
architectures. The aim of these experiments is to desing the
agent's controllers by defining a set of behavioural states and a
set of probabilistic transition rules between the states which
allow the agents to aggregate. Since at this stage of the project,
s-bots were still in the prototyping phase, we carried out our
work by mainly using simulated agents. However, some experiments
have been run on real robots built with LEGO Mindstorm©.
The work on simulation focused on the study of three
shapes of possible aggregates:
- Cluster
pattern formation
- This is the most simple form of aggregate. The behaviours
used for it are the base for the all the other aggregations.
- Chain pattern
formation
- The s-bots have to form chains. This structure can be
useful, for instance, if pulling of heavy objects is required.
- Centre/periphery
pattern formation
- This structure consider two kinds of s-bots. They have to
cluster, but with s-bots of one kind in the centre and the other
outside. It can be useful if, for instance, s-bots in the centre can
carry heavy objects and those outside have more sophisticated sensors
to look for the best path.
The work with real robots focused on chain formation: <>
Chain formation
using LEGO Mindstorm©.
The methodological approach mentioned above (i.e., behavior-based
approach) have been
subsequently abandoned. The work on aggregation has continued by
exploiting artifcial neural networks shaped by evolutionary
algoithms. In particular, evolution has been used to automatically
design the control system of the s-bots for aggregation in a cluster.
This work on scalable
aggregation behaviours is described in the following:
Evolving
Clustering Formation
Control >> Aggregation
| 
