Kangaroo - traction test

Comment

Comment by Andy Payne on June 3, 2012 at 9:47am

Very nice!  Congrats.

Comment by Ángel Linares on June 3, 2012 at 9:21am

That's it!

Comment by Daniel Piker on June 3, 2012 at 8:25am

I guess you mean this : http://boxcar2d.com/*

and yes, this was one of the inspirations for the little demo above.

I haven't tried yet, but it should be possible to evolve some vehicle designs.

A couple of examples of Galapagos+Kangaroo : 1 , 2

* which uses the awesome Box2D engine by Erin Catto, which is also the engine behind Crayon Physics and Angry Birds - a funny story about this here

Comment by Ángel Linares on June 3, 2012 at 8:07am

Mmm...I saw some months ago a video about a evolutionary car solver writed in Processing, Cynder, Open Frameworks or something like that. That solver measured the success of the "car" driven over the same terrain with little jumps, ramps, etc.

The was transformed mutating his features: distance between wheels, shape of the "nose", speed,...

If you let the solver running for few minutes, finally you get a great and fast car design. This example that you show us remember me a lot that solver...perhaps mixing galapagos and kangaroo...

Comment by Ángel Linares on June 3, 2012 at 7:54am

Nice improvements! Thanks for this amazing work Daniel.

Comment by Daniel Piker on June 3, 2012 at 7:41am

To explain a little more :

In Kangaroo, when a particle moves below the floor level, its position gets projected back to the surface, and its speed in the vertical direction gets reversed, and multiplied by some restitution factor between 0 and 1. This affects how high it bounces.

By default the horizontal component of a particle's velocity is unaffected by floor collisions, so a bouncing ball will keep moving horizontally.

However, if you want to push against the floor to alter your horizontal movement (i.e. traction), then this is no use, as feet or tires will just keep slipping without imparting any acceleration.

This is why I added the 'tumble' setting - which controls how much of the horizontal component of a particle's velocity gets retained in a collision with the floor. With it set to 0.5, the above simulation turns into this:

and here is a comparison of the effect on a box :

The latest change (to be available in the next release) is to add this option for mesh collisions also - so instead of vertical and horizontal, we look at the components of the velocity of the colliding particle in the directions normal and tangent to the mesh respectively.

I'd be interested to hear if anyone has any ideas of how they might want to use this...