Unexpected physics issues - box collider missing a corner

[ottworks]

Here are the issues I'm experiencing: https://user-images.githubusercontent.com/5253130/136295955-51616d9b-be93-4e0e-b5de-3a589bba6006.mp4

Here is the source code: tetra.zip

[jmiskovic]

I notice there are differences in the ways our drawCollider functions render the shapes. My version seems to match the underlying physics simulation. You could try to switch between the two and see differences in collider orientations.

-- call lovr.math.setRandomSeed(0) before looping over colliders in lovr.draw()

function drawCollider(collider)
  local userData = collider:getUserData()
  local shade = 0.4 + 0.4 * lovr.math.random()
  lovr.graphics.setColor(shade, shade, shade)
  local shape = collider:getShapes()[1]
  if false and shape:isSensor() then
    local r,g,b = lovr.graphics.getColor()
    lovr.graphics.setColor(r,g,b,0.8)
  end
  for _, shape in ipairs(collider:getShapes()) do
    local shapeType = shape:getType()
    local pose = mat4(collider:getPose()):translate(shape:getPosition()):rotate(shape:getOrientation())
    local x,y,z, _,_,_, angle, ax,ay,az = pose:unpack()
    -- draw primitive at collider's position with correct dimensions
    if shapeType == 'box' and shape then
      local sx, sy, sz = shape:getDimensions()
      lovr.graphics.box('fill', x,y,z, sx,sy,sz, angle, ax,ay,az)
    elseif shapeType == 'sphere' then
      local r = shape:getRadius()
      lovr.graphics.sphere(x,y,z, r)
    elseif shapeType == 'cylinder' then
      local l, r = shape:getLength(), shape:getRadius()
      lovr.graphics.cylinder(x,y,z, l, angle, ax,ay,az, r, r, true, 10)
    elseif shapeType == 'capsule' then
      local l, r = shape:getLength(), shape:getRadius()
      local m = mat4(x,y,z, 1,1,1, angle, ax,ay,az)
      lovr.graphics.cylinder(x,y,z, l, angle, ax,ay,az, r, r, false)
      lovr.graphics.sphere(vec3(m:mul(0, 0,  l/2)), r)
      lovr.graphics.sphere(vec3(m:mul(0, 0, -l/2)), r)
    end
  end
end

I haven't checked your manual calculation of mass and inertia tensor, that will become obsolete once the #490 is solved.

Then, if I increase the collider mass to 10, the physics seem to work smoother although still not correctly. This is most likely because the inertia tensor is not correctly computed from all the shapes.

We should probably merge this issue with #490.

[ottworks]

Thanks for the updated drawCollider code, this one works better with lighting. I've increased masses all around, and it does seem like the inertia tensor is incorrect, physical reactions to collisions are in the wrong direction (sometimes).

https://user-images.githubusercontent.com/5253130/137645871-ffc805d8-b357-431e-b218-61dd99960dcb.mp4

[ottworks]

Oops! setAngularDamping(0.01) fixed it! My sphere colliders were apparently spinning very rapidly. The only remaining problem is that my table is missing a corner.

[bjornbytes]

LÖVR uses Jolt instead of ODE now. The mass APIs are improved so this should be fixed.