Interface with external visualizer

[diegoferigo]

JAXsim for the time being does not include any GUI nor visualizer to show the simulated models. While prototyping, I used gazebosim/gz-gui part of Gazebo Sim, but I think it's a dependency too large that would make deployment quite challenging.

It would be quite interesting using gazebosim/gz-gui as standalone visualizer, but I never saw anything based on it.

Alternatively, rdeits/meshcat-python could be a valid and slim alternative, but in the past I couldn't manage to obtain acceptable performance in real-time usage. The problem is that all nodes (link poses) had to be refreshed individually. However, few months ago, duburcqa/jiminy#543 implemented a method to update all nodes together, and it's worth a try.

[diegoferigo]

I created ami-iit/meshcat-viz-python to play with MeshCat, hoping to achieve better real-time performance by optimizing the client/server protocol that streams link transforms.

It would be also useful (but with much lower priority) being able to tweak the control menu of the web interface. I opened meshcat-dev/meshcat-python#128 for asking implementation details.

[flferretti]

As discussed F2F, the code relative to the visualization loop that allows to visualize the contact forces using meshcat-viz can be found in ami-iit/element_rl-for-codesign/.../hopper.py:

Visualization Loop

for t in range(len(x_t.base_position[0]))[::downsampling]:
    for env_id in range(len(model_names)):
        world.update_model(
            model_name=model_names[env_id],
            joint_names=model.joint_names(),
            joint_positions=x_t.joint_positions[env_id][t],
            base_position=x_t.base_position[env_id][t],
            base_quaternion=x_t.base_quaternion[env_id][t],
        )

        # Compute the forces expressed in the right frame for visualization
        CL_fc_ext = world_to_viz_contact_forces(
            model,
            x_t.base_position[env_id][t],
            x_t.base_quaternion[env_id][t],
            x_t.joint_positions[env_id][t],
            time_history[model.name()].aux["t0"]["contact_forces_points"][env_id][t],
        )[:, 0:3]

        # Update the visualized contact forces
        for idx, (CL_fc, name) in enumerate(zip(CL_fc_ext, parent_link_name)):
            contacts[env_id][f"{name}_{idx}"].force.set(force=np.array(CL_fc))

        rate.sleep()

[diegoferigo]

I guess that we eventually converged towards the usage of the passive Mujoco viewer in #83. Closing.