Code center > Darwinbots3

New bot shape

<< < (2/15) > >>

fulizer:
yes but try to avoid I being to power comsuming for the computers

Numsgil:
It shouldn't take that much extra computational power.  The bots will always be convex, which makes collision detection not that much more difficult.

And as long as I'm posting, currently I'm envisioning a bot that can control it's length, width of the head, and width of the tail.  So I bot could squish itself down to a circle, or squeeze its head and tail and form a really long noodle.  Should allow sodaplay-esque multibots to form, but with more control over individual elements.

shvarz:
What if you don't have flexible joints at all, but rather put much more focus on flexible shapes? Two bots would define how much surface they would share with each other and which part of them would be shared. Then they are tightly connected there, but are free to change their own shapes. This would be much more biologically relevant. Would that work at all or would it be way to complicated?

EricL:

--- Quote from: shvarz ---What if you don't have flexible joints at all, but rather put much more focus on flexible shapes? Two bots would define how much surface they would share with each other and which part of them would be shared. Then they are tightly connected there, but are free to change their own shapes. This would be much more biologically relevant. Would that work at all or would it be way to complicated?
--- End quote ---
It would be more computationally expensive.  The more you can describe a shape with a formula (e.g. circle, line, rectangle) the faster it will be to draw, the faster we can compute collisions and so on.  The more we allow arbitrary shapes, the more work has to get done make up those arbitrary shapes from lots of smaller geometric shapes (usually triangles).  It gets costly.  Special hardware can help (which is what graphics cards and Xboxes do) but you still have to compute the shapes and collisions and so on.

Some strategies such as using bounding boxes can help and may provide realistic looking collisions, etc. by using a simplier invisible shape that surrounds the more complex arbitrary shape.  That can help narrow down the cases where the work has to be doen for collision detection and so on, but we still have to do the work when collisions do occur.   For arbitrary shapes to matter, those tiny little surface differences have to matter and make a difference in how organisms collide, etc.  

Bottom line, simple shapes make the math easier and sims faster but also the organisms simipler.  You get what you pay for.

Numsgil:
I explored floppy bean bag like cells, but I had these problems:

1.  Most systems like this aren't really that robust.  You can have the cell collapse in on itself in a tightly coupled mass of vibrating goo.  Basically game over the cell.  Which is fine for a game where you only lose a life and start the game over, but for an evo sim it's a disaster.

2.  Floppy shapes are not always convex, making collision detection significantly more difficult.  We'd probably never get sims with more than a few hundred bots run at any decent speed.

3.  I have no idea how a bot would control its shape from the DNA.  Actual cells use a sort of fuzzy logic, pulling various bits of its cytoskeleton with varying strength, which results in the final shape.  I don't really have a good idea of how to accomplish something similar with the way DNA works in the program.

Still, if those three issues ever get solved, I'd really like to explore floppy ameoba cells.  It's just beyond my capabilites (and most others' capabilities from the looks of it).

Navigation

[0] Message Index

[#] Next page

[*] Previous page