UMI-Underwater: Learning Underwater Manipulation without Underwater Teleoperation

1. 1Collect on-land human demonstrations of grasping
2. 2Transfer grasp knowledge to underwater domain using depth-based affordance representation
3. 3Autonomously collect successful underwater grasp demonstrations via self-supervised data collection pipeline
4. 4Fine-tune the system for improved performance in underwater manipulation

Robotics engineers and AI researchers on a team can benefit from this system as it enables autonomous collection of underwater grasp demonstrations and transfer of knowledge from on-land to underwater domains, improving the efficiency of underwater robotic grasping

💡 Transfer of grasp knowledge from on-land human demonstrations to underwater domain is possible using depth-based affordance representation

🤖 UMI-Underwater: learning underwater manipulation without underwater teleoperation! 💡