Agibot G2 Data Aquisition Kit

The Agibot G2 Data Acquisition Kit is a specialized hardware and software bundle designed to bridge the gap between human demonstration and autonomous robotic intelligence. Tailored for the AgiBot G2 “Genie” wheeled humanoid, this kit serves as the primary gateway for “Embodied AI” training—allowing developers to collect high-fidelity multimodal data to train the robot’s GO-1/GO-2 LLM (Vision-Language-Latent-Action) models.

(Price on request)

Core Components & Features

The kit is built on a UMI-isomorphic layout (Universal Manipulation Interface), ensuring that the data collected by a human operator matches the robot’s physical configuration perfectly.

Teleoperation Hardware: Includes high-precision controllers (often VR-based or exoskeleton-style) that enable 1:1 motion replication. It captures human arm and hand movements with sub-millimeter accuracy.
MEgo Perception System:
- MEgo View: A head-mounted or robot-mounted vision suite that captures egocentric video data aligned with the robot’s perspective.
- MEgo Gripper: Specialized data-collection grippers that record tactile feedback and grasping forces during a demonstration.
Genie Studio Integration: A full-stack development platform that automates the processing, reconstruction, and annotation of the collected data, turning raw demonstrations into training-ready datasets.

Key Specifications

Feature	Description
Precision	Force control accuracy $\le$ 0.5N for delicate tasks.
Data Modalities	Synchronized Vision (RGB-D), Motion (DoF tracking), and Tactile data.
Compute Support	Optimized for the NVIDIA Jetson Thor platform (2070 TFLOPS).
Architecture	Isomorphic mapping to the G2’s 7-DoF arms and 12-DoF OmniHand.
Latency	Ultra-low latency for Beyond-Visual-Line-of-Sight (BVLOS) control.

Why It Matters

Traditional robot programming is rigid. The G2 Data Acquisition Kit enables Learning from Demonstration (LfD). Instead of writing code for a specific task (like sorting a specific industrial part), a human wears the kit, performs the task once, and the robot “learns” the spatial relationships and force requirements.

Industrial Applications

Precision Assembly: Teaching the robot to handle flexible cables or sub-millimeter electronic components.
Complex Logistics: Data collection for irregular object sorting and kitting.
R&D & Research: Serving as a foundation for labs developing new Embodied AI models using the AgiBot World open-source dataset.

Note: While the G2 robot itself is an industrial powerhouse with a 185 kg frame and IP42 rating, this Data Acquisition Kit is the specific “teaching tool” that allows it to evolve from a pre-programmed machine into a truly intelligent, task-aware collaborator.