openworm
0.1
Collaborate to generate a clean worm.cc model to start working from. It seems that both @mstachowsky and @portegys have been working on the code separately. These two models are probably incompatible by this point. @mstachowsky's code has been used for various experiments and can be abandoned.
After a bit of digging, it seems that this will require us to agree on a set of behaviours that we are interested in. It appears from the literature that certain behaviours are only possible due to certain neural properties. If we do not include these properties in the core system, then the simulator/robot cannot perform them.
I think we have to be resigned to using only a subset of the connectome, just as most pure software model do. I think the salt-seeking/foraging behavior would be pretty novel for a robot and I have something of a handle on a model from the literature. Of course touch is a close second for me.
I agree. To make things concrete, here is my suggestion:
- forward locomotion
- backward locomotion
- independent head/neck movements from the body
This repertoire would give us almost everything the worm can do, with the exceptions of some of the more extreme turns it's capable of. However, missing from this is the important question of magnitude of response. Touch responses at least involve larger magnitudes for stronger touch. How exactly this works is a complete unknown (that is, to the research community) but there are some suggestions for how it might arise. If we ignore this, then we may need to re-write a major part of the core code. If we don't, then we're working with unknowns. I'm open to thoughts...