The Year of CoCoRo 38/52: After we focussed in the last 2 months on showing quite complicated search scenarios we will now focus for some weeks on the topic of "collective awareness" and "collective decision making". In this early laboratory experiment shown here we implemented the BEECLUST algorithm in a swarm of Lily robots. The BEECLUST is a simple swarm algorithm derived from the walking and resting behavior of young honeybees which can compare several temperature spots in their environment and collectively choose the optimal (warmest) spot. Thus the group is aware of all options in its environment and chooses the best one without having every single individual agent knowing all available options. In the example we show in the video here the algorithm was "translated" from honeybees to underwater robots as follows: All robots move randomly in the habitat. When they meet another robot, and only then, they measure how deep the water below them is. The more shallow it is, the longer they stay in place. By running this algorithm, the swarm is able to identify shallow places and collectively choose the shallowest of all. Of course the same can be done to find the deepest point, the darkest or the brightest point in the habitat, it is only one more step necessary: correlating the resting time of the robots with other local environmental properties. The BEECLUST is one of the simplest swarm algorithms possible, maybe it is even THE simplest algorithm that you can do and still call it a "swarm algorithm". Simplicity is very nice, as it allows easy translation from one domain into another domain. However, an algorithm like this does not suit for every application, as we had to learn: In our experiments we found that, in contrast to crawling honeybees in the hive or driving wheeled robots on the ground, it is very tricky for an AUV to stay in place in water because of drift due to water turbulences. Even in an aquarium this is already an issue if many robots move around and induce such turbulences. The algorithm relies on this rest-in-place behavior, so we concluded that for a more turbulent underwater habitat there have to be better algorithms than the classical BEECLUST.
Социальные закладки