ALOHA 2

Google DeepMind's improved, open-source, low-cost bimanual teleoperation hardware, ALOHA's second generation.

ALOHA 2 is open-source bimanual teleoperation hardware released in February 2024 and posted to arXiv in May of that year, led by an “ALOHA 2 Team” at Google DeepMind that included researchers such as Chelsea Finn, and is an improved version of Stanford's original ALOHA. It keeps the same low-cost “move the leader arm by hand, the follower arm copies it” approach, with the main improvements focused on ergonomics and durability: the force needed to operate the leader arm's gripper dropped to about a tenth of before, the follower arm's grip strength doubled, and the data-collection software runs on ROS 2, logging data at 50 Hz. The full hardware design is open-sourced, along with a system-identified MuJoCo simulation model. The bimanual data behind both ALOHA Unleashed and Gemini Robotics was collected on this kind of platform.

ExampleThe folding-paper and lunchbox-packing bimanual tasks demonstrated by Gemini Robotics were performed on an ALOHA 2.

Also called

ALOHA2

Related

ALOHA · Mobile ALOHA · ALOHA Unleashed · Leader-Follower Teleoperation · Google DeepMind · Gemini Robotics

Sources

arXiv 2405.02292: ALOHA 2: An Enhanced Low-Cost Hardware for Bimanual Teleoperation

As of

2024-05