Talk:Evolutionary robotics
| WikiProject Robotics | (Rated B-class, Mid-importance) | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|||||||||||||||||
I have a few problems with the statement Robot controllers typically consist of artificial neural networks, and evolution typically creates the controllers by modifying the strengths of connections between the neurons in the neural network.
The first part, "Robot controllers typically consist of artificial neural networks" is blatantly false when taken by itself. A revision of the article might want to be careful not to imply that *all* or *most* robotic controllers are based on artificial neural networks.
The second part may very well be true, but the evolutionary robotics work which I am the most familiar with does not use artificial neural networks at all (see: http://www.aic.nrl.navy.mil/~schultz/papers/robolearn.96/robolearn.96.html)
It might be more informative to mention a suite of techniques which are common in evolutionary robotics such as genetic algorithms (see http://mitpress.mit.edu/catalog/item/default.asp?ttype=2&tid=3684)
Contents |
[edit] Page is missing works on evolution of morphology
This page ignores some works in evolution of morphology, such as Lipson and Pollack's GOLEM, Genobots by Hornby, Molecubes, recent Fab@Home movement towards printing of robots, Framsticks project, Tom Ray's Virtual Pets, etc.
[edit] MDS -- Changes made dec 01
Cleaned up grammar, clarified some things. Rewrote misleading statement implying that robot controllers in all contexts were usually neural networks.
Discuss evolutionary robotics methods which do not use neural networks.
Added section explaining the motivation for evolutionary robotics from a robot learning perspective.
Article may now be far too focused on the machine learning applications of evolutionary robotics.
I maintain that it still needs expert attention.
[edit] Why was this taken out?
The first couple revisions had the opening
"A field of robotics (or computer science, or theoretical biology, depending on your point of view) where controllers for (physical or simulated) robots are created through artificial evolution, e.g. genetic algorithms. Less often the physical setup of the robots is itself evolved."
I really liked these two sentences (note that I am not the person who wrote them) and believe the article should have some mention of evolutionary algorithms used to design the physical setup of a robot. Since I am largely unfamiliar with these experiments (I do know of one minor one that was proposed) I reccomend that an expert add a section referring to the highlights of this area.
[edit] Origins of the field
I have deleted
| “ | The term evolutionary robotics was introduced in 1993 by Cliff, Harvey and Husbands at the University of Sussex [citation needed]. | ” |
The field has been around for longer than that - didn't Koza do evolution of genetic programs that controlled a Logo turtle in the eighties? I'm not sure where the exact term originated, but this claim has been unreferenced for two years now, time to delete. I'm pretty sure that the statements about '92 and '93 being the years of first evolved controllers is wrong too, for the reasons already mentioned. It is also unreferenced. 78.105.234.140 (talk) 11:48, 1 November 2009 (UTC)
