Darwinbots Forum
Code center => Suggestions => Topic started by: Numsgil on October 23, 2006, 07:11:58 AM
-
I'm working on The New Tie Paradigm (http://www.darwinbots.com/WikiManual/index.php?title=New_tie_paradigm), specifically the unfinished bits: angular constraints and muscles. I'm looking for feedback...
Key Point: ties aren't springy, they're rigid. Like a steel rod.
1. Imagine 3 bots are ties to each other forming a triangle. Imagine that the requested lengths of the 3 ties is 50, 100, and 200.
Obviously this is an impossible triangle. 200 > 50 + 100. The physics code will resolve this in a stable way but the exact end result is going to depend on the bots' position in the robot array, which is bad, but not a deal breaker.
If anyone has a good idea for a more equitable solution to this problem, that would be great. I'm not sure searching all ties to see what circuits form is a good idea (note that this problem could be expanded to many, many, bots).
Whatever occurrs, it needs to be stable, ie: predicable and constant from cycle to cycle.
2. Angular constraints are when bots "freeze" the ball bearing joint between themselves or other ties. For instance, a bot might freeze itself to a tie so that it always points in a certain direction (this is how ties work now), or it might freeze the angle of two other ties (to keep the resulting structure rigid).
Bots can "unfreeze" a joint at any time during their cycle.
I'm thinking of implementing these such that:
- You can't request an angle, you can only freeze an existing angle. If your ties are 60 degrees apart, you can't "freeze" them at 30 degrees unless you move them some how. This prevents bots using angular constraints to generate motion for free or little energy (they should use tie muscles for this).
- Angular constraints are "soft", meaning that they are updated at the beginning of the physics code, and aren't checked again after bots move. Basically collisions need to be 100% (or nearly so) every cycle so the bots look "solid", but other hard constraints don't need to be as perfect, so they don't need to be worked on as hard. Tie length checks occur after angular constraints, so angular constraints take a back seat to requests to change the length of a tie if there is a conflict between the two.
- Conceptually, an angular constraint places a tie between the two bots on the other end of both ties, moving the bots so that the resulting angle between their ties fits the constraint.
3. Muscles - this one has me stumped. The bots act as joints, the ties act like bones. The middle bot needs to somehow give an order that can apply forces to the ties to move them.
Basically, a squid bot needs to be practical. A bot tied to two other bots that forms a traingle without a base. The middle bot would constrict the ties together to create a forwards moving force.
Can someone point me to a diagram of where and how muscles produce torque on bones in real life?
-
1a. Add properties for tie tensile strength for compression and tension. Make tie tensile strength a function of the nrg used to create it. Calculate the forces applied on ties by bot movements including acceleraton due to tie operations. Break ties where the force on them exceed their tensile stength.
1b. Add the concept of a tie to nowhere, a spike. Give tie creation / length change operations have a propagation rate. Tie lengths can only be changed by n per cycle. Tie creation operations happen over the course of several cycles as a function of how far away the tied bot is.
2. Do away with tie angles all together. See #3 below.
3. Make it possible for ties to attach to/from bot edges, not just to bot centers. Use two or more ties for muscle/bone operations or for angular positioning. For a muscle/bone, have one tie from bot center to center, and from edge to edge with the attachment points seperated by 180 degrees. Change the lengths on the edge conneted ties to move the muscles...