View RSS Feed

Development

  1. Step one complete, now for step two

    by
    Aperion
    , 02-08-13 at 05:26 AM
    This is a follow up post to my previous one about refactoring and my plan forward.

    I have completed the first item in my list
    Way Forward (aka the plan)
    1. Replace slidenode code with outside drive-train code - Completed
    2. Refactor current wheel calculation - in progress
    3. Rewrite current drive-train to make use of new code
    4. Add new drive train features


    Here is what I wrote about the next step here:
    Refactor current wheel calculation
    Current wheels aren't very reliable, we average the speed of each node to calculate the rotation of the wheel, this is subject to fluctuation. Instead I will have one object that represents the wheel as a whole, it acts more like a traditional tire model in a game where it's simply a cylinder. This is very similar to slidenodes is that there is an ideal state, and a state the wheels actually are at.

    I'm expecting this to help solve a few issues RoR currently faces such as vehicles slipping down a slop when they should be braked. Also under this approach folding wheels should be a thing of the past.
    I am kind of gathering my thoughts on how to approach this step, and how to explain it. There are essentially two "worlds" I'll be dealing with, a beam world and a drive-train world. the beam world is the one we are familiar with, point masses held together by springs in a 3D space. The drive-train world is made of point masses held together by springs in a 1D space. So where nodes have an [x,y,z] coordinates drive-train components simply have an angle. Essentially the drive-train world is the beam world with different dimensions and units.

    So what the next step needs is translating from 1D to 3D and back. The angle in the drivetrain world will correspond to the angle offset each node is from it's original position, the hard part is how do you account for the wheel being at any odd angle? Possibly even upside down. Come to think of it, I remember now what my solution was, this is actually why I write these posts, to help me think.

    What we know is each nodes initial starting position and the current position of the axle nodes and each wheel node. If we find the rotation required to translate the axle to it's original postion we can then apply that transformation to the current wheel node position. Then use the translated current wheel node position, the center axle and the initial node position we can calculate the current angle difference from the initial position. This can then be used to calculate difference between where the node is supposed to be based on the drive-train world, and where it actually is.

    I did write a tiny program just to test this out:
    Code:
    #include <iostream>
    #include <OGRE/Ogre.h>
    
    int main()
    {
    
    	Ogre::Vector3 pn1(1, 0, 0);
    	Ogre::Vector3 pn3(1, 0, 1);
    	
    	Ogre::Vector3 n2(0, 0, 0);
    	Ogre::Vector3 n1(0, 1, 0);
    	Ogre::Vector3 n3(-1, 1, 1);
    	Ogre::Quaternion rot;
    
    	 // move nodes to origin...
    	n1 -= n2;
    	n3 -= n2;
    	
    	// get rotation
    	rot = n1.getRotationTo(pn1.normalisedCopy());
    	
    	// move spoke to origin for rotation
    	n3 -= n1;
    	n3 = rot * n3;
    	
    	std::cout << Ogre::Degree((pn3 - pn1).angleBetween(n3));
    	return 0;
    }
    We'll see how logn this takes to get done. night night.
    Categories
    Development
  2. Refactoring slidenodes, drivetrain, and framework

    by
    Aperion
    , 01-28-13 at 10:53 PM
    Since I've been doing some coding I thought I'd share what I'm working on. First and foremost my goal right now is to rework the drive-train code, this includes the tires. But I am sort of taking a round about approach to this which includes some pretty heavy refactoring. I've written a lot of concept code in the past for a new drive-train but never put it to use. I also wrote it in a reusable manner and am expanding this design.

    History
    The drive-train code I did write works on it's own, it just needed to be integrated into RoR. The new drive train code is built on the same principle as RoR, nodes and beams. Instead of nodes and beam in a 3 dimensional space we have nodes and beams in a 1 dimensional space. replace position with angle, velocity with moment, and force with torque and you have the same equations. This is where I started.

    So I am basically building a parallel "world" for the drive train that works similar to nodes and beams. My starting point was going to be the wheels, since they essentially act like the gate way between the node/beam world and the drive-train world. My approach here is to simulate a drive-train wheel object which is nothing more than an an angle, then calculate where the nodes should be positioned based on that angle. While I was working on this, I realized that this is drive-train same concept I'm using for slide nodes: calculate where a node should be, calculate beam-like forces from where the node is, and where the node should be and distribute forces appropriately. So I started hacking away at some concept that I could use for both when I dawned on me that I had already written this code a long time ago for the drive train.

    Theory
    So here is a bit about my theory behind the drive-train code, and what will most likely be used in many many more places than the drive-train. As I've worked in the BeamForcesEuler routine I've realized that it boils down to two actions, updating the forces that act on a node, and updating the nodes position. I call them sources, and sinks; sources generate/calculate the forces based on the environment and are applied to the sinks, the sinks take the forces and decide what to do with them. In RoR terms sources are beams, engine, wings, commands, collisions, etc, etc, and sinks are nodes.

    One of the cool ideas I will be able to introduce with this new idea is per-node integrators, I feel this will be handy because some weight/spring combinations are not stable but would be if a different integration method is used.

    Way Forward (aka the plan)
    1. Replace slidenode code with outside drive-train code
    2. Refactor current wheel calculation
    3. Rewrite current drive-train to make use of new code
    4. Add new drive train features


    Replace slidenode code with outside drive-train code
    I'm doing this first because it's an existing system so I will be able to just test out the new code against existing slide node behavior to tell if it is working. This will be the biggest change all other items will make use of the concepts implemented here.

    Refactoring this code will also have the added benefit of slide node damping since it's already built into the code I will be integrating.

    Refactor current wheel calculation
    Current wheels aren't very reliable, we average the speed of each node to calculate the rotation of the wheel, this is subject to fluctuation. Instead I will have one object that represents the wheel as a whole, it acts more like a traditional tire model in a game where it's simply a cylinder. This is very similar to slidenodes is that there is an ideal state, and a state the wheels actually are at.

    I'm expecting this to help solve a few issues RoR currently faces such as vehicles slipping down a slop when they should be braked. Also under this approach folding wheels should be a thing of the past.

    Rewrite current drive-train to make use of new code
    Make sure the current drive train code works and behaves as is with the new wheel model, and make use of new code to perform drive train calculations. The current code is a bit of a mess and there is a bug where the first two axles are locked together but none others, this should solve that problem.

    Add new drive train features
    Multiple engines, torque converters, clutches, lockers, limited slip, multiple gear boxes, transfer-case, brakes, etc

    My goal here is to model any and all kinds of different power, be it a traditional engine and 4 wheels, a parallel hybrid with two different engine types, or a helicopter with one engine powering two different rotors. Or who knows what. I want to flip over a car, spin one wheel with the mouse and watch the other wheel spin in the opposite direction when using an open diff. the best part IMO is that I've already got some of this working. I've test locked, open, and LS diffs. I've tested with one power source, and I've tested with two power sources.

    And Beyond!
    I plan on using this framework throughout RoR, hoping to achieve better performance, multiprocessor support, more organized, and easier to alter.

    I'm excited

    My work and progress is viewable here: http://sourceforge.net/u/aperion/rig...0fd7a443/tree/
    What I will be using for much of my refactoring is here: http://sourceforge.net/u/aperion/rig...ics/framework/
    Categories
    Development
  3. Hi

    by
    Aperion
    , 12-13-10 at 06:59 PM
    Hello world, been thinking about RoR lately, there is so much to do that sometimes I can't help but think the easiest thing would be to start over.

    Since I've had some idead on my mind I'll write them down or else I will forgot about them:

    • Beam structure and architecture:
      • Currently a list of beams and nodes is stored with each vehicle, I wonder if it would make more sense to store ALL beams and nodes in some sort of central storage.
        Potential Advantages:
        1. One large continuous memory instead of in multiple places
        2. Threading: no artificial barriers on how many beams are processed by a thread, a single truck could be processed by multiple threads
      • Object Orientation: Break up the physics into two type of classes, sources, and sinks. Sources are the physical objects that forces can be applied to, and sources and the physical phenomenon that generates forces. In the current sim, nodes would be the only kind of sink. But the sources would be just about anything else, beams, engine, wings, buoyancy, slidenodes, etc. possibly even collisions
    • server
      • Move the physics to the server to allow proper calculations for collisions. if each client is doing local collision calc then there is probably too much time between updates. This means a truck would move too far and generate too large of a force. Only problem is this dramatically increases the load on the server, and might not make it feasible.


    Ok I've run out of steam, but those are my thoughts for now.
    Categories
    Development


About Rigs of Rods

    Rigs of Rods is a unique soft body physics simulator.


Some Tools


Partners

SourceForge.net

Follow us

Twitter youtube Facebook RSS Feed


impressum