GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture
GigaBrain-0.7 reports better zero-shot and post-training robot task performance after scaling to 37,000-plus hours of embodied data.
The paper presents a three-system architecture meant to unify understanding, prediction, and action for vision-language-action robots. It says the model improves over earlier GigaBrain-0 releases and prior systems including pi_0.5. The reported gains cover language-conditioned instruction following and task success across home and industrial scenarios, including the team’s Maker H01 platform and mainstream robot embodiments. Training code and pretrained weights are promised for release. HF Daily Papers' note
The paper presents a three-system architecture meant to unify understanding, prediction, and action for vision-language-action robots. It says the model improves over earlier GigaBrain-0 releases and prior systems including pi_0.5. The reported gains cover language-conditioned instruction following and task success across home and industrial scenarios, including the team’s Maker H01 platform and mainstream robot embodiments. Training code and pretrained weights are promised for release. HF Daily Papers' note
score 6