pymunk | use pythonic 2d physics library that can be | Animation library

The big downside is performance. It is much more expensive to call step 10000 times than 1 time for a given period of time.

I imagine that sooner or later there will be other effects as well, for example around the floating point precision when the dt becomes very small. E.g if an object has a very low velocity it will move a tiny distance, and if that number is very small it might not be possible to accurately store it as a floating point number (64 bit double in the case of pymunk), so that errors accumulate and the end result is a worse simulation than calling step only once.

Im dont know enough of the collision algo to answer from the mathematical background, but I can give some help on the way:

1. This is from the collision algorithm used. For easy (e.g. Circle to circle) collisions only one point is found, as you can see here: https://github.com/slembcke/Chipmunk2D/blob/master/src/cpCollision.c#L526 In more advanced cases, e.g. Poly to Poly, then the GJK & EPA algorithms are used, https://github.com/slembcke/Chipmunk2D/blob/master/src/cpCollision.c#L607 and this can produce one or two points.

2. I think the reasoning is shown in the below picture. The points are located on each shape, and since they can be overlapping it becomes natural with one for each.

1. I suspect so, but am not 100% sure.

If you are intersted about the GJK & EPA implementation, there's a intro blog post from when it was first introduced into Chipmunk here: http://howlingmoonsoftware.com/wordpress/enhanced-collision-algorithms-for-chipmunk-6-2/

StaticPoly.bodies contains pymunk.Bodys. These bodies must also be added to the space.

Reed the error message:

The shape's body must be added to the space before (or at the same time) as the shape.

Either before

The 2nd argument of pymunk.Circle is the radius, but not the diameter:

shape = pymunk.Circle(body, 60)

Either you can use the static body already attached to the Space, or you make a new one. To create a rectangle, either you provide the corners to the Poly constructor (as shown below), or you use the shorthand Poly.create_box method.

The problem is that both Poly shapes that are created with the shorthand create_box method are created with their center at the position of the body they are attached to. So the two boxes are positioned on top of each other.

To fix this you need to use the more generic Poly constructor. In this way you can pick the positions as you want.

If I just take your picture and write up the coordinates times 10, it would be something like this:

1. The torque is reset every step, as written here: http://www.pymunk.org/en/latest/pymunk.html#pymunk.Body.torque Given this I think the expectation is that it should slow down over time. I should also add that usually its a bit tricky to set these kinds of values manually, better would be to use a SimpleMotor constraint.

2. Yes, the center of gravity seems to be the issue here just as you write. The center of gravity will be at 0,0 of the Segment (and similar at 0,0 in case you create a Poly shape). So c1 has the center of gravity in the middle, while c2 has it at one end. You can either just change the definition of c2 to be (b2, (0,-20), (0,20)...) and modify the position of the body to compensate, or use the body center_of_gravity property to adjust its location. http://www.pymunk.org/en/latest/pymunk.html#pymunk.Body.center_of_gravity

3. I think the issue is that the moment you use for the two segment bodies are very extreme. The moment can be calculated with the built in method moment_for_segment, e.g.: pymunk.moment_for_segment(1000, (0, -40), (0, 40), 6) The calculated moment is 717000, which is quite different from the value in your code. You can also let Pymunk calculate the mass and moment for a body from the Shapes attached, http://www.pymunk.org/en/latest/pymunk.html#pymunk.Body.__init__

4. Pymunk uses the C library Chipmunk for the actual simulations. The focus of this library was mainly gaming or other realtime uses, and the same for Pymunk. However, I know that Pymunk has been used for many different purposes, as shown in the list of uses here: http://www.pymunk.org/en/latest/showcase.html page. If the algorithms inside Pymunk are accurate and comprehensive enough for your use case I do not know.

5. Not sure.

A final note is that its probably a good idea to use as small dt in the space.step function as possible since that will produce better simulations.

You have to draw the objects before updating the display

I am getting an error message: