UMI-Underwater: Learning Underwater Manipulation without Underwater Teleoperation

2026-03-27

AI Summary

This work addresses the challenges of underwater robotic grasping, including severe image degradation, drastic lighting variations, and the absence of underwater human demonstration data. The authors propose a cross-domain grasping framework that requires no underwater teleoperated demonstrations. By leveraging self-supervised collection of successful underwater grasp samples, the method transfers grasping knowledge from terrestrial human demonstrations to underwater environments through depth-based affordance representations, a geometric alignment mechanism, and an affordance-conditioned diffusion policy, enabling zero-shot deployment and policy fine-tuning. Experimental results demonstrate that the approach significantly improves grasping success rates and background robustness in pool environments, generalizes to objects seen only in terrestrial data, and outperforms RGB-only baseline methods.

Abstract

Underwater robotic grasping is difficult due to degraded, highly variable imagery and the expense of collecting diverse underwater demonstrations. We introduce a system that (i) autonomously collects successful underwater grasp demonstrations via a self-supervised data collection pipeline and (ii) transfers grasp knowledge from on-land human demonstrations through a depth-based affordance representation that bridges the on-land-to-underwater domain gap and is robust to lighting and color shift. An affordance model trained on on-land handheld demonstrations is deployed underwater zero-shot via geometric alignment, and an affordance-conditioned diffusion policy is then trained on underwater demonstrations to generate control actions. In pool experiments, our approach improves grasping performance and robustness to background shifts, and enables generalization to objects seen only in on-land data, outperforming RGB-only baselines. Code, videos, and additional results are available at https://umi-under-water.github.io.

Problem

• underwater robotic grasping
• domain gap
• visual degradation
• demonstration scarcity
• affordance transfer

Innovation

• underwater manipulation
• self-supervised learning
• domain transfer
• affordance representation
• diffusion policy