Humanoid Single-Hand G1+Dex3 / Inspire hand Grasp Dataset | RGB-D IMU Contact | Offline RL - Robotics Data
Overview
Humanoid Single-Hand G1+Dex3 / Inspire hand Grasp Dataset | RGB-D IMU Contact | Offline RL - Robotics Data is a video dataset. Typical applications include robotics & embodied AI. Relevant sectors: AI & Machine Learning, Education and Manufacturing & Industrial. Pricing is not published and is provided on request.
Primary applications
The provider Humanoid Robot G1+Dex3 Single‐Hand Grasping Dataset This is a robot data product ready‐to‐use for model training. All expert‐level single‐hand manipulation trajectories are generated on our in‐house simulation platform built primarily upon NVIDIA Isaac Sim and Isaac Lab. The hardware setup consists of the Unitree G1 humanoid robot equipped with one Dex3 three‐fingered dexterous hand, executing a complete workflow of approach, grasp, transport and placement. Each data episode covers only one type of grasping motion; no mixed hand movements are present within a single demonstration sample. The single‐hand payload is limited to 0.5 kg or less. The G1 model is calibrated against real Unitree hardware: mass, inertia, joint limits and link dimensions are derived from physical measurement and hardware calibration, facilitating sim‐to‐real transfer of trained policies. This dataset includes four categories of single‐hand grasping motions: - Pinch‐Side: Two vertical sides of the object are clamped by three fingers with top‐down enclosing grasp. Approach pose: open dexterous hand approaches vertically from above the cube. Placement is performed with slow aligned lowering; fingers open only after the object makes contact with the supporting surface. - Power‐Envelop: The dexterous hand wraps around the mid‐section of the object. Grasp position is 30‐40 mm below the lower edge of the object opening; grasping on object edges is prohibited. Approach pose: spread dexterous hand move
Technical specifications
- Modalities
- Video
Related datasets
More in Video & Physical AI →Multiple Scenes Ego-Centric Data | Egocentric Data | Embodied Data | 100K Hours
1. Overview 100K Hours Multiple Scenes Ego-Centric Data, human manipulation data captured from a first-person view across real-world scenarios.
- Language
- —
- Volume
- 100k h
- Domain
- AI & Machine Learning +2
- Use case
- Robotics & embodied AI