Robot-Use LLM Agents
VLA/VLM/LLM Data Collection
Data collection for VLA, VLM, and LLM training: the hardware, the scenes, and the cleaning.
A model learns to control a robot from data. The most useful source is probably robot data itself. After that, the next most useful may be game data, 3D data, video data, and code.
We handle data collection end to end: we design the capture hardware, design and build the scenes, and clean the data, including removing sensitive material.
We can supply varied scenes from our own studio, with professional set construction and diverse layouts. We also provide first-person recordings of people working in biology laboratories, chemistry laboratories, and chemical plants.
Branches
Robot Data
Probably the most useful kind, and what LLMs are and will be trained on.
Game, 3D, Video, and Code Data
The next most useful sources.
Collection Hardware
Hardware designed for capturing the data.
Scene Design
A studio, professional set building, and varied layouts, so scenes change often.
Data Cleaning
Removing sensitive material before data leaves us.
First-Person Lab and Plant Data
Human operation recorded in first person in biology labs, chemistry labs, and chemical plants.