Oceanveo provides high‑quality, structured training datasets of human hand‑object interactions captured by a global network of data collectors. The data include multi‑modal signals such as fingertip pressure, contact geometry, and hand kinematics, annotated across interaction phases to improve robot perception and manipulation. These ready‑to‑use datasets help robotics manufacturers and AI labs train more reliable models for real‑world deployment.
Funding
Funding not disclosed
Founders
Product
Problem
Robotic systems often struggle to perform reliably in unstructured, real‑world environments because existing training datasets are limited to generic, simulated, or lab‑captured data that lack the diversity of human hand‑object interactions. This gap leads to brittle perception and manipulation capabilities when robots are deployed in warehouses, homes, hospitals, or factories.
Solution
Oceanveo addresses this gap by deploying a global network of human data collectors who record purpose‑built interaction scenarios, capturing detailed information such as fingertip pressure, surface contact, dual‑hand coordination, reach phases, and multi‑finger grip dynamics. The collected data are structured and annotated to serve as high‑quality training material for AI models that control robotic manipulators. By providing datasets that reflect real‑world human perception and dexterity, Oceanveo enables robots to achieve more accurate grasping, object stabilization, and task execution across diverse physical settings. The company’s pipeline delivers ready‑to‑use data that can be integrated into existing machine‑learning workflows, accelerating the development of robust, production‑grade robotic solutions.
Target Audience
Primary customers are robotics manufacturers, AI research labs, and automation solution providers that need realistic manipulation data to train and validate perception and control models for real‑world deployments.
Features
- Global human data collection network capturing multi‑modal interaction signals (pressure, contact geometry, hand kinematics)
- Structured annotation of hand‑object interaction phases: approach, reach, grip, and stabilization
- High‑resolution, purpose‑built scenarios covering a wide range of objects and tasks relevant to industrial and service robots
- Dataset formats compatible with common robotics and deep‑learning frameworks (ROS, TensorFlow, PyTorch)
- Quality control processes ensuring consistency, diversity, and repeatability across collected samples