Page 9, Pixel Observations code.
Argument to pixel Wrapper should be pixels_only, not pixel_only.

Changing an integrator's accuracy after initializing a TimeStepper has no effect, but this does not seem to be documented.

Discovered in opensim-org/opensim-core#2041.

With flares, you can set a custom activation key using a control number. The control numbers range from 0-500, and 1 would be CTRL+1, 2 would be CTRL+2 and so on, makes sense. However, that page also lists control numbers for other functions:

-   `11` Clutch
-   `12` Parking Brake
-   `40` Starter
-   `45` Axle L

This may be a lack of grokking Chipmunk, but I noticed that calling Space.add_collision_handler with *(a, b) and *(b, a) gave different handlers, but only one's separate handler was ever called.

Is the ordering significant? I tried swapping the order that I added collision handlers, but it always picked the same one (ie. not just the last one defined).

This isn't a complete sample

Currently the output at runtime produces, for instance, the following:

PIConGPUVerbose PHYSICS(1) | Courant c*dt <= 1.66765 ? 1
PIConGPUVerbose PHYSICS(1) | species e: omega_p * dt <= 0.1 ? 3.56797e-08

The way the output is formatted I interpret it as a "question to the code", i.e., "Is c*dt <= ?" To which a natural answer would be "Yes/No" or "True/False" and not

We should add solref and solimp arguments to MujocoGeneratedObject and its subclasses to easily play around and experiment with contact modeling. The default behavior for some objects is pretty bad - for example thin cylinders tend to sink into the table.

It would be helpful to add a tutorial explaining how to run things in parallel (mpirun etc). There are some users who are not familiar with mpi (see #173 ) and it is useful to explain the basic steps to do this.

The boosted-frame example script (in the documentation) uses the direct laser injection.
However, when modifying parameters, users can easily be in a situation where the laser does not fit in the box.
For this reason, it might be better to use the laser antenna in the default example.

Using, for example, napoleon, the Python API docs could be improved to better list parameters and their types.

Note there seems to be an issue with numpy return types: https://stackoverflow.com/q/56620817/4069571