VLA-Δ: Learning Physical State Transition for Vision-Language-Action Models
VLA-Δ trains robot policies on physical state transitions and beats π0.5 in real-robot tests.
VLA-Δ is a vision-language-action framework that treats physical changes between current and future states as explicit learning targets, rather than mapping observations directly to demonstrated actions. The authors build a large pretraining dataset with annotations for task-relevant semantics and visual motion. At inference the model generates actions from current observations without extra overhead. It is competitive with the state of the art on LIBERO and RoboTwin and outperforms π0.5 in real-robot experiments.
- VLA-Δ supervises physical state transitions between current and future states.
- A large pretraining set annotates task semantics and visual motion changes.
- Inference emits actions from current observations with no extra overhead.
- It matches state of the art on LIBERO and RoboTwin and beats π0.5 on real robots.
- Organizations
- ZJU-OmniAI
From the original article
1 min read at huggingface.co
Vision-language-action (VLA) models transfer semantic knowledge acquired through vision-language pretraining to robot control, achieving strong performance across various benchmarks. However, most VLAs directly map observations to demonstrated actions without explicit supervision of intermediate physical state changes, making them prone to overfitting observation-action correspondences.
By Qi Lu, Jiarun Li, Ce Zhang · 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.