SLA Model Scales Spatial Understanding via Continual Pre-training
Original title: SLA: Scaling Spatial-Language-Action Model via Continual Pre-training
SLA optimizes action prediction and spatial alignment from frozen 3D perception for robotic manipulation.
SLA is a Spatial-Language-Action Model that scales implicit spatial alignment through continual pre-training initialized from a pretrained pi05 policy. It jointly optimizes action prediction and alignment between intermediate visual embeddings and spatial representations. The model uses feature-dependent gates and probabilistic state dropout while reusing pretrained VLA weights for inference.
- Curates and unifies heterogeneous robot datasets
- Feature-dependent gates adaptively aggregate visual embeddings
- Probabilistic state dropout mitigates overfitting to trajectories
- All auxiliary components used only during training
From the original article
1 min read at huggingface.co
Reliable robotic manipulation requires both semantic understanding and precise control in three-dimensional space. However, vision-language-action (VLA) foundations largely inherit visual representations from 2D vision-language pre-training, while spatial supervision is often introduced only during task-specific adaptation.
By Fuhao Li, Wenxuan Song, Qizhou Huang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.