open-gigaai2026apache-2.0

GigaBrain-0.7

An embodied foundation model with a three-system architecture (understanding, prediction, action) trained on over 37,000 hours of heterogeneous embodied data, achieving improved generalization across diverse robot embodiments and tasks.

Downloads174
Hours37,000
Likes4

Why This Matters for Physical AI

GigaBrain-0.7 demonstrates how scaling embodied foundation models with heterogeneous data and improved architectural design enables emergent cross-embodiment generalization capabilities critical for developing practical robotic agents across diverse real-world scenarios.

Technical Profile

Modalities
rgblanguage
Robot Embodiments
Maker H01multiple heterogeneous embodiments
Action Space
vision-language-action
Environment
homeindustrial
Task Types
manipulationcomplex_taskslong_horizon_taskslanguage_conditioned_instruction_following
Total Hours
37,000
Annotation Types
language_instructions
License
apache-2.0
Part of the GigaBrain family

Community Signals

Access

Need custom rgb data?

Claru builds purpose-built datasets for home applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets