inertia box fluid model #191
Conversation
mujoco_warp/_src/passive.py
Outdated
| lvel_torque = rotT @ torque | ||
| lvel_force = rotT @ force | ||
|
|
||
| if m.opt.wind[0] or m.opt.wind[1] or m.opt.wind[2]: |
There was a problem hiding this comment.
not sure if there is a big benefit to this, but could be wp.max(m.opt.wind) > 0.0.
Or I guess this information is even statically available, so maybe wrap in wp.static? actually, you're not running that code right now anyway if all of them are zero, are you going to reuse this function in a different kernel?
There was a problem hiding this comment.
wrapped in wp.static
There was a problem hiding this comment.
the functionality from _inertia_box_fluid_model is moved into box_fluid
| @@ -227,22 +227,6 @@ def test_get_data_into_m(self): | |||
| np.testing.assert_allclose(mjd.qLD, mjd_ref.qLD) | |||
| np.testing.assert_allclose(mjd.qM, mjd_ref.qM) | |||
|
|
|||
| def test_option_physical_constants(self): | |||
There was a problem hiding this comment.
maybe we should still check for ellipsoid fluid and fail?
| opt_viscosity = opt.viscosity | ||
| opt_density = opt.density | ||
|
|
||
| @nested_kernel |
There was a problem hiding this comment.
This gets a bit subjective, but I think the bar might be higher for using a nested kernel. For tiles we have no choice, but for here, we can pass the opts in as params.
If it led to a significant performance difference I would understand, but this kernel looks pretty cheap.
In particular, my understanding is that the performance hit comes from thread divergence. e.g. some threads have viscosity and others don't. I don't think that's the case here, no?
adds inertia box fluid model
https://mujoco.readthedocs.io/en/latest/computation/fluid.html#inertia-model
https://github.com/google-deepmind/mujoco/blob/d3bc0544d39c0933067e493f4c1c7ebfd15ae9f3/src/engine/engine_passive.c#L527