EgoMimic Hardware Repository

This repository contains the hardware setup for EgoMimic: Scaling Imitation Learning via Egocentric Videos. To train imitation learning policies, or to collect human embodiment data via Aria glasses, see the main EgoMimic repo.

Useful Links

• Project Page
• Hardware Assembly Document

---

Meet Eve

Eve is the robot behind EgoMimic, built to learn from Egocentric Videos Effortlessly.

This codebase interfaces with all of Eve's hardware components, including the ROS setup to control ViperX arms and support Aria Glasses integration in ROS. It also contains scripts for collecting teleoperated demonstrations on Eve. We build off of the Aloha ROS2 implementation from Trossen Robotics.

---

Software Support

• ✅ Supported Operating System: Ubuntu 22.04
• ✅ ROS Version: ROS 2 Humble

Structure

• eve: Python package providing useful classes and constants for teleoperation and dataset collection. Contains scripts to use Aria glasses with ROS node (stream_aria_ros.py)
• hardware: Contains CAD designs for all EgoMimic hardware, including Aria glasses mounts, Arm mounts, and gripper.
• config: a config for each robot, designating the port they should bind to, more details in quick start guide.
• launch: a ROS 2 launch file for all cameras and manipulators.
• scripts: Python scripts for teleop and data collection

Install instructions

Build all of this without a conda environment in ~/interbotix_ws

1. Set up ROS2, interbotix packages, and realsense cameras via the (Trossen Instructions). But critically under the step for Aloha Software Installation, make sure to clone this repo to '~/interbotix_ws/src'instead

IGNORE: git clone https://github.com/Interbotix/aloha.git
git clone https://github.com/SimarKareer/EgoMimic-Eve.git

2. Setup Aria Glasses Aria Instructions. In brief you just need to run

python3 -m pip install projectaria_client_sdk --no-cache-dir
aria-doctor
aria auth pair # (then open mobile app and approve)

3. Install extra packages

pip install opencv-python fastplotlib==0.1.0a13 glfw==2.6.2 pygfx==0.1.14 dm_env h5py ipython

4. Add permissions on video cams

sudo usermod -a <user> -G video

5. Add aliases to ~/.bash_aliases

alias setup_eve="source /opt/ros/humble/setup.bash && source ~/interbotix_ws/install/setup.bash"
alias ros_eve="ros2 launch eve aloha_bringup.launch.py"

6. Restart machine

7. If facing issues, build realsense ros package from scratch, I used this commit

Tips

• Use a standard terminal instead of vscode.
• Any time changes are made to the ROS package, you must run colcon build in ~/interbotix_ws. (As of December 24, 2024, there are 39 packages to build at this stage.)
• ros2 run rqt_reconfigure rqt_reconfigure

Usage

To start up ROS run

setup_eve
ros_eve # (wait 15 seconds for aria to start streaming)

To ensure all camera sensors are functional launch Rviz

rviz2

Occasionally Aria fails to connect, simply restart the computer and it should connect fine.

Data Collection / Teleop Scripts

For data collection and teleop we'll continue to use the base python installation (no conda env)

Teleoperation

python scripts/dual_side_teleop.py --arm <left, right, both>

Data Collection

First, define a task in eve/constants.py. After adding this you'll need to rebuild via colcon build. Then run

python scripts/dual_side_teleop.py --arm <left, right, both> --task_name <task name>