Self-replicating (macro-)bots

A brief communication in this week’s Nature reports (macroscopic) machines that autonomously self-replicate from randomly positioned components. The work is from Saul Griffith and colleagues at MIT’s Media lab. Self-replication is done in a two stage process; there’s a recognition step, in which when two correctly oriented building blocks randomly collide they latch together, and there’s an error correction step, in which incorrectly joined sub-units are separated.

A movie (9 MB Quicktime movie – I think this works without a subscription) shows the self-replication happen. The blocks are placed on an air-table and agitated to bring blocks randomly into contact with each other. Of course, this is just a macroscopic analogue of the random, Brownian motion that is so important at the nanoscale. It’s interesting to compare this with another, much publicised, example of a macroscale, self-replicating system reported earlier this year in Nature. In that case, self-replication was a deterministic process that relied on its components being supplied in a well-ordered way. Griffith’s approach (see his web-site for more context) consciously mimics the self-assembly processes used in biological systems; the architecture of the structure is encoded in each of its components, and assembly depends on random interactions between these components. The combination of these two features is what leads to the huge potential advantage of this approach to self-replication over more deterministic techniques – the potential to make the process massively parallel and inherently scalable. This is fascinating and thought-provoking work.