ULOHA: An Underwater Bimanual Robot System for Robot Learning

Summary

Underwater visuomotor policy learning has focused primarily on single manipulators, while bimanual imitation learning has been studied largely in air.

Abstract Summary

Underwater visuomotor policy learning has focused primarily on single manipulators, while bimanual imitation learning has been studied largely in air. We present ULOHA, an underwater bimanual robot learning platform that combines custom-designed leader--follower hardware with software extensions to LeRobot, integrating teleoperation, multi-view sensing, demonstration collection, policy training, and autonomous deployment. Real-robot experiments demonstrate a range of coordinated underwater bimanual behaviors, including inter-arm transfer, shared-object manipulation, and buoyancy-driven interception. We evaluate ACT, Diffusion Policy, and the vision--language--action model SmolVLA on the platform. We investigate how learning methods and execution strategies developed for manipulation in air perform underwater, examining bubble disturbances, buoyancy-driven object motion, action-execution horizons, and real-time chunking. A separate single-arm study examines policy transfer between air and water and shows that demonstrations spanning both media support execution in both under the tested conditions. ULOHA provides a unified experimental platform for studying underwater bimanual robot learning under the coupled perceptual and physical effects of underwater environments. Additional material: https://mertcookimg.github.io/uloha/

Key Points

Underwater visuomotor policy learning has focused primarily on single manipulators, while bimanual imitation learning has been studied largely in air.
We present ULOHA, an underwater bimanual robot learning platform that combines custom-designed leader--follower hardware with software extensions to LeRobot, integrating teleoperation, multi-view sensing, demonstration collection, policy training, and autonomous deployment.
Real-robot experiments demonstrate a range of coordinated underwater bimanual behaviors, including inter-arm transfer, shared-object manipulation, and buoyancy-driven interception.
We evaluate ACT, Diffusion Policy, and the vision--language--action model SmolVLA on the platform
We investigate how learning methods and execution strategies developed for manipulation in air perform underwater, examining bubble disturbances, buoyancy-driven object motion, action-execution horizons, and real-time chunking.

ULOHA SYSTEM

ULOHA combines hardware designed by the authors with a software stack that extends LeRobot [ 11 ] through ULOHA-specific hardware adapters and recording, training, and deployment configurations. The hardware comprises two dry leader arms and two kinematically matched underwater followers, with custom arm structures and grippers.

EXPERIMENTAL EVALUATION

The evaluation addresses four questions: which coordinated underwater tasks can be learned from demonstrations; whether ACT, Diffusion Policy, and SmolVLA can be deployed on the platform; how bubbles and action-execution strategies affect deployment; and whether policies transfer between air and water.

We first evaluate bimanual learning and deployment, then use a separate single-arm ACT study to examine training-medium coverage.

EXPERIMENTAL EVALUATION — ULOHA: An Underwater Bimanual Robot System for Robot Learning — Fig. 5: Views from the right hand camera during demonstrations of buoyant blue cellulose sponge manipulation. (a) Surface hand-over: the right gripper grasps a floating sponge and transfers it to the left gripper. (b) Release and catch: the left gripper releases the sponge, and the right gripper intercepts it as it rises.

Sources and Demonstrations

Figures are reproduced from Kobayashi et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.