Jon Simpson

Robotic Cockroaches, again.

18 Nov 2004 — behavioural, bio-inspired, robotics, occam-pi, cscs

On my earlier theme of Cockroaches and potentially of interest to those following the current stream of CSCS, is a story from Slashdot regarding the use of robotic cockroaches, whereby real cockroach societies control the behaviour of a group.

The end result of a cockroach that can successfully integrate into societies and influence them is the product of a $3.3 million program, perhaps slightly above the capabilities of CSCS, but still interesting and relevant to the work that we’re currently doing. Interesting phrases I picked from the target article include:

Cockroaches like contact with each other. When they meet, they stay still. They are happy to be with a friend for a few moments. The more friends around them, the longer they stay.

When it bumps into a cockroach, it does what they do: it stops moving. The more cockroaches that approach it, the longer it remains stationary.

There is also material that seems to feed into my earlier post on the hierarchy of processes that have successively lesser significance to the being, something Matt picked up and developed upon (giving me additional insight and thoughts on the issue) over at the CSCS weblog. The material may suggest that if we can get the robots to communicate and sense each others’ presence, we can add a social layer to their behaviour that fits into the architecture of our cockroach.

His experiments place cockroaches in a space that contains two shelters - one dark, one light. Naturally, they gather in the dark shelter, where they feel comfortable. But if the robots go to the light shelter, cockroaches follow - the desire for companionship proving stronger than the need for dark.

The fact that accurately modelling cockroach behaviour to the point where it can fool a real cockroach took 3 years of research is interesting, and it will be a good challenge to see how far we can get in the remaining 4 weeks of CSCS with our similar project. Unfortunately the website of “Leurre” , the group responsible for this search is currently under development and doesn’t seem to contain much in the way of useful documentation on the behavioural model they have implemented within the robot, something I was hoping to be able to examine and use whilst developing on the Mindstorms. There are papers they have published on the subject, and I may try to follow these up if possible (time & resources suggest otherwise).