This paper presents the effect of a simple vibration message passed locally between two robots on the success of the whole swarm in implementing a complex best-of-N manipulation task. The task is called the generalized leaf curling task. On an unknown leaf containing edges with multiple levels of stiffness, a group of very simple robots is required to find and collaboratively curl up one of the softest edges. In earlier work, using the "relative-value based randomness" method [Phan and Russell, 2010], our robots have demonstrated the ability to complete this task. However, the success of that algorithm depends strongly on the working environment and requires parameters to be preset. In this work, by incorporating the transfer of a simple message between two robots via local vibrations, the whole robot swarm is able to explore the environment in a particular way that results in finding better and better objects over time. With this trend, it is conjectured that, given enough time, the robots will find the best object in the environment. The success of the new algorithm which is called "local maximum conservation" is demonstrated via high completion rates of the robot swarm in different complex working environments. The algorithm was developed using physical robots and verified by a series of tests using a visualized simulator.
展开▼
