Self-assembly

One of the first major simulations to be performed has been self-assembly, based on principles derived in the work of Liu and Winfield [Distributed autonomous morphogenesis in a self-assembling robotic system, Morphogenetic Engineering: Understanding Complex Systems, 2012].

We've got our self-assembly strategy working for Multiple Entry Recruitment to produce shapes defined on the seed robot. This strategy makes use of local line-of-sight infrared communications and docking guidance cones, plus a small amount of Wi-Fi communications to transfer initial data.

Below a series of videos display self-assembly into a variety of structures:

 

Furthermore our strategy is able to operate with a moving seed robot:

There are some improvements still to be performed to the self-assembly algorithm, namely improving the ability of robots involved in docking to warn others away and prevent collisions, but this should help provide an initial baseline.

These simulations were made in V-REP, those with the most robots manage to run at around 1/3 to 1/4 of realtime speed on an Ubuntu PC with a 3.2GHz intel i5 CPU and 8Gb of RAM. The smaller simulations can be made to run a little faster than realtime.