Abstract
In this paper, we present a system that will be able to obtain microscopic assembly behaviours for a robotic swarm to achieve an assembly target (macroscopic model). It will be designed taking into consideration the essential features of a self-assembling system needed to be implemented in a real robotic swarm. This system is composed of a typology of generative languages PD0L, and an algorithm for generating individual rules to be processed by the robots. The assembly process will be performed in a distributed manner, and will be also designed to require minimal communication capabilities between robots. Both the expressive capacities of language and the rule generation algorithm will be demonstrated by evaluating their performance with a core set of test morphologies widely used in self-assembly tasks. Furthermore, we compare the assembly time and the number of messages required between a classic controller (centralised) and our distributed approach.
Keywords:
Acknowledgements
This work has been supported by the Ministerio de Ciencia e Innovacion of Spain, project TIN2009-10581
Notes
We must take into account that if we deal with a homogeneous swarm, and the robots do not have mechanisms to detect a broken link (either because a robot stop working or disconnects) then the system will not be able to complete the assembly regardless of the algorithm used. However, if the robots can detect a broken link, it is easy to adapt the LSDA algorithm to repair the assembly: a disassemble signal can be sent to all the robots that connect with the broken link (these robots are known because they belong to a branch that is determined when executing the corresponding rule of the L-System) and then restart the assembly process in this branch. Even if more robots than available are needed to complete the assembly (once discarded the faulty ones), the final structure could minimise the most important failures using a well-designed L-System, as discussed in Section 5.2.3.