// the find
MarkFzp/mobile-aloha
Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation
This is the hardware bring-up and teleoperation code behind the Mobile ALOHA paper: a low-cost bimanual mobile manipulation rig built on Interbotix arms and an AgileX Tracer base. It's for robotics researchers trying to reproduce the demo or collect their own imitation-learning datasets, not a general-purpose manipulation library.
The udev symlink instructions for pinning arms and cameras to stable device names (ttyDXL_master_right, CAM_HIGH, etc.) solve a real and annoying ROS problem where USB device paths shuffle on reboot. The scripts map cleanly onto the actual data pipeline you need for imitation learning: record_episodes, visualize_episodes, replay_episodes. It's the reference implementation behind an influential, widely-cited demo, so it's a solid starting point if you're building the exact same rig rather than reverse-engineering a paper.
Hard-coded to specific hardware (Interbotix ViperX/WidowX arms, AgileX Tracer base) and ROS1 Noetic on Ubuntu 18.04/20.04 only — ROS2 and newer Ubuntu are listed as 'ongoing' but never shipped before the repo went quiet. This repo alone doesn't train anything; you need the separate ACT-for-Mobile-ALOHA repo to actually run imitation learning on the collected data. No CI, no automated tests, no packaging — it's a hand-assembled script collection with manual steps (Dynamixel Wizard current-limit tweaks, editing a vendored arm.py to skip FK) rather than something you pip install and trust. Last commit was mid-2024, so any dependency drift (ROS, camera SDKs, AgileX's pyagxrobots) is on you to fix.