Taylor Howell

This is a great task, thanks for the contribution! Things to improve performance: - shorten the planning horizon: `` - change the integrator to `implicitfast`: `` - modify the model...

Have you tried this task with the G1 model?

Yes, it looks like we are missing this file. @erez-tom is this file created from a Menagerie model?

The state is set below in this [line](https://github.com/google-deepmind/mujoco_mpc/blob/1f78168912b4fe4af30365198c9f4e3ecf025ba1/mjpc/app.cc#L382). The planner has access to the most up-to-date state. However, I am noticing that we might not be initializing the planner state...

@lrchit which task produces this error?

- ellipsoidcylinder - cylindermesh - ellipsoidhfield are available in MJX via the MuJoCo Warp backend (set `impl='warp'`) ``` mx = mjx.put_model(m, impl='warp') dx = mjx.put_data(m, impl='warp') ```

The documentation for MJX feature parity is updated to note that all flex features are unsupported. Additionally, `io.put_model` is updated to check for flex and raise errors if the `mjModel`...

@giorgionicola adhesion actuators are implemented for MuJoCo Warp for example, simulating `mujoco/model/adhesion/active_adhesion.xml` with MuJoCo Warp's viewer ``` mjwarp-viewer --mjcf=mujoco/model/adhesion/activate_adhesion.xml ```

This feature is available in MJX via the MuJoCo Warp backend by setting `impl='warp'` ``` mx = mjx.put_model(m, impl='warp') dx = mjx.put_data(m, impl='warp') ```

it is expected that the python api introduces additional overhead compared to directly running the c++ code. however, depending on the particular application this additional overhead might or might not...