Host: The Japan Society of Mechanical Engineers
Name : Dynamics and Design Conference 2019
Date : August 27, 2019 - August 30, 2019
A cooperative transportation system of different-size objects to each corresponding destination has been developed as an application of Swarm Intelligence. It is well known that several kinds of insects may achieve complex tasks in a group, where each insect may play its role by the information of its surroundings. By making a swarm, the insects achieve difficult tasks that an individual cannot manage alone. This collective behavior, as emerged in a group of social insects, has been called Swarm Intelligence. Although the Swarm Intelligence has been applied to control a robotic swarm, these artificial systems still depend on an outside datacenter to share the information or on a controller placed outside of a swarm to get rules for each robot. In order to improve the system and implement the natural behavior into the artificial control system, a swarm should have the learning ability that may correct and acquire its behavior by itself. In this study, a cooperative transportation task of different-size objects to the proper destinations by lots of workers has been simulated. Transportation goals are varied to the size of objects. Smaller objects should be carried one goal and larger objects should be carried to the other goal. In addition, workers could not sense the size of the object solely. Workers are required to estimate the size of objects based on positional information of their neighbors. A solution of the task of decision-making problems about a transportation goal has been proposed by using an artificial neural network. Considering individual workers in a group as a unit in the artificial neural network, the learning algorithm of neural networks has been applied to decide the transportation destinations. Each worker may decide his action based on nearby objects and the state of nearby individuals, and the swarm transports objects under the control of the swarm network. As a result, it is shown that the swarm acquires the learning ability how to decide on an appropriate goal for given objects.
