A Comparison of Three Insect Inspired Locomotion Controllers

A Comparison of Three Insect Inspired Locomotion Controllers Cynthia Ferrell, of the Brooks Lab. The paper was published sometime in the mid nineties

She compares a spectrum of approaches from particularly decentralized 'reflexive' based controllers (Beer, Chiel and Cruse) to somewhat the more centralized 'Patterned' controllers (which use less sensation and more central coordination of walking patterns) (Brooks, Collins and Stuart), with a look at 'Hybrid' control schemes incorporating elements of both. I should say now that, as of two weeks ago, I am a student of Beer.

I was excited about the paper because I thought it would offer general insights into the relative advantages and disadvantages of each approach. HOwever, it was mostly a chronicle of the implementation of each model on their robot, Hannibal, along with a review of the models the work is based on. This betrays their interest in the practical, which, for starters, isn't what I was looking for.

When it comes down to it, Ferrell's results on Hannibal favor patterned approaches to insect robot locomotion.

Mention is made of the Ambler (Krotkov, Simmons and also Thorpe, '90, '92), which seems like a model for extremely centralized approaches to control.

Two important quotes: "Another drawback of reflexive controllers is their difficulty to add new behaviors. The controller mechanisms interact in a very specialziezed way to produce gait coordination. Consequently is it difficult to incorporate dramatically different behaviors without disrupting controller dynamics" and, conversely: "In contrast, the appeal of a patterneed approach is the ease of adding new features to the system" I have to get my head around this observation. Is this a reflexion of their platform or a fundamental aspect of centralized control that new behaviors are easier to add. Does it reflect a lack of a procedure in the science of distributed approaches for adding new behaviors? Also, what are behaviors?

Questions:

• It sounded like implementation on a robot designed for centralized locomotion models interfered with the ability of the researchers to characterize the full pros and cons of each approach. In a few places they describe problems they encountered implementing the reflexive approaches at the pure 'reflexive' researchers didn't. In general, there seemed to be a good amount of fudging from the 'ideal' of each model to actual implementation. This is understandable, but perhaps could be controlled better in the future.

• If there was a debate here, it happened a good decade+ ago. Where is the controversy now?

To what extent were biologically inspired approaches implemented on either of the Martian rovers. Why not?

• Much of the biological inspiration for all of the models comes from Wilson '66. Is this E.O.?