|
Control >> Cooperative transport >> Different shapes and sizes
Transport of objects of different shapes and sizes
Aiming at a cooperative transport behaviour which is less
sensitive to
the characteristics of the prey than the ones described previously, we conducted a new experiment
with the following focus: the objective is to let the s-bots
self-assemble into structures which are capable of pulling or pushing
prey of different shape and size towards a target location.
In this study, the simulated s-bots are provided with the
ability to
rotate their turret with respect to the chassis, In addition, the
s-bots can sense each other with the camera.
Experimental setup
Four s-bots are put at random positions in the neighbourhood
of the
prey. The s-bot controllers are supposed to let the s-bots localise
and approach the prey, and self-assemble into structures physically
linked to the prey, in order to pull or push it towards a beacon.
The prey is modelled as a cylinder or as a cuboid of height 12
or 20
cm. Depending on its weight, the cooperative behaviour of at least 2
or 3 s-bots is necessary to move the prey.
The s-bot can control its left and right wheels, the rigid
gripper,
and the orientation of the chassis with respect to the turret. In
addition, the s-bot can control the heading of the (simulated)
directional camera.
Using its directional camera, the s-bot perceives its
teammates and
the prey. It can also determine the direction to a beacon (the target
location), unless it is shadowed by an object. In addition, the s-bot
can determine whether it is grasped to another object or not.
The group of s-bots is controlled by a simple recurrent neural
network
that is synthesised by an evolutionary algorithm. All the s-bots of a
group transporting a prey are initially equipped with an identical
neural network. To favour the evolution of controllers that make use
of the gripper element, the fitness function takes the assembling
performance into account.
An evolutionary algorithm is utilised in order to obtain the
weights
of the s-bots' neural controllers. Starting from a population of 80
random initial candidate solutions, each one is assigned a fitness
value reflecting its quality. In each generation, the best 20
individuals are selected to produce the subsequent generation
of 80 candidate solutions.
Results
The experimental setup described above has been used in ten
independent evolutionary runs of 850 generations each. The evolved
controllers perform quite well, independently of the shape and size of
the prey, and allow the group to transport the prey towards a moving
target. Additionally, the controllers evolved for a relatively small
group can be applied to larger groups, making possible the transport
of heavier prey.
References
- Groß R. and Dorigo M. Cooperative Transport of Objects of Different Shapes and Sizes, In M. Dorigo, M. Birattari, C. Blum, L. M. Gambardella, F. Mondada, and T. Stützle, editors, Ant Colony Optimization and Swarm Intelligence, 4th International Workshop (ANTS 2004), volume 3172 of Lecture Notes in Computer Science,
Springer Verlag, Berlin, Germany (2004) 107-118
Control >> Cooperative transport >> Different shapes and sizes
| 
