Aloha Workspace for Simulation

[j3soon]

This PR adds support for the following features:

1. Robotic arm (Aloha arm)
2. MoveIt 2
3. Isaac Sim

These features were not present in the original codebase and can serve as references for future workspaces that incorporate robotic arms, MoveIt 2, or Isaac Sim.

Closes: #48

Isaac Sim

There are 3 ways to install Isaac Sim:

1. Install through Omniverse Launcher (i.e., binary install)
   • https://docs.omniverse.nvidia.com/isaacsim/latest/installation/install_workstation.html
   • https://isaac-sim.github.io/IsaacLab/source/setup/installation/binaries_installation.html
2. Use the pre-built Isaac Sim docker image
   • https://docs.omniverse.nvidia.com/isaacsim/latest/installation/install_container.html
   • https://isaac-sim.github.io/IsaacLab/source/deployment/docker.html
3. (Experimental) Pip installation
   • https://docs.omniverse.nvidia.com/isaacsim/latest/installation/install_python.html
   • https://isaac-sim.github.io/IsaacLab/source/setup/installation/pip_installation.html

The first option is challenging to set up within a container. The second option involves using a closed-source Dockerfile, which requires a free NGC account to pull the image. This approach adds difficulties for future maintenance for several reasons:

• It locks us into a specific Docker base image while using Isaac Sim, limiting control over important configurations like the Ubuntu version.
• ROS 2 must be manually installed on top of this image, leading to larger image sizes and preventing the reuse of existing images, such as osrf/ros:humble-desktop-full.
• Upgrades can become difficult, especially if closed-source changes in the base image cause package compatibility issues.
• If a normal user is used, need to manually chown all isaac-related files
• and more...

The third option, while still experimental, simplifies Isaac Sim installation through pip install, allowing for easier integration. Although not all tutorials are updated for this method, I have come up with straightforward workarounds and can make Isaac Sim integration similar to that of Gazebo.

In this PR, I choose to use the third option, and believe it can allows much easier future maintenance for this repo. Nevertheless, it is worth noting that the first two options remain valuable for scenarios where long-term maintenance isn't necessary, such as quick testing or prototyping.

Limitations

• Currently, only simulation is supported (as documented in the README) because I don't have access to the physical hardware yet.
• Currently, it only supports x86 machines (also documented in the README).
  • I plan to test on AGX Orin in the near future and perform corresponding changes.
• Some hardcoded solutions are in use, such as ExecuteProcess used in the launch file for Isaac Sim.
  • This will be fixed in another PR dedicated for the isaac sim workspace. (as in issue #47)

In addition, after this PR is merged, we'll need to update template_ws and other workspaces in a subsequent PR to ensure full Isaac Sim support across all workspaces. The modifications in the aloha_ws also fixes several issues present in the current template_ws.

Would appreciate if @YuZhong-Chen or @Assume-Zhan have some time to review this and provide some feedback. If you're able to, please add yourself to the Reviewers section before beginning the review. Thank you!

[j3soon]

Tested on a clean machine with L40 GPUs. If host UID isn't 1000, the permission issue can be resolved by this simple workaround: https://github.com/j3soon/ros2-essentials/issues/53#issuecomment-2351397799

[j3soon]

Note that this PR is tested with docker engine version 27. Lower docker versions such as version 25 will report error due to the subpath feature used.

Ref: https://docs.docker.com/engine/release-notes/26.0/

[j3soon]

Just tested on a Ubuntu 22.04 Desktop environment with ARM CPU. MoveIt 2 can work without any code changes.

Please note that Gazebo and Isaac Sim do not support ARM CPUs, so trying to run them on such hardware will fail as expected.

[j3soon]

When controlling the Isaac Sim robot arm with MoveIt 2, it often exhibits overshooting behavior compared to Gazebo. The cause of this issue is still unclear.

I suspect this issue may be related to the dampness and stiffness when importing the URDF. However, the dampness seems to be explicitly set to 0.1 in the vx300s URDF. The following are some potential references:

• https://docs.omniverse.nvidia.com/isaacsim/latest/advanced_tutorials/tutorial_advanced_import_urdf.html#importing-urdf-using-python
• https://docs.omniverse.nvidia.com/isaacsim/latest/features/robots_simulation/ext_omni_isaac_gain_tuner.html#options-for-tuning-gains
• https://forums.developer.nvidia.com/t/how-can-i-set-stiffness-and-damping-with-urdf-importer/287362/5

I think we can document this behavior in README and open an issue. This issue could be left to future investigation when we are integrating our UR5 arm.

[j3soon]

I'll go ahead and merge this first. If you notice any issues in the future, please add your review here, and we can address them with a follow-up PR. Thanks!