VLA Model
Vision-Language-Action model, an important paradigm in robot learning that transfers the representational capabilities of large vision-language models (VLMs) to robot control tasks, leveraging internet-scale pretraining knowledge to improve policy generalization.
Core Facts
Timeline (last 90 days)
VLA models rely excessively on the camera configuration at training time, and once the number or poses of cameras change in the deployment environment, model performance may drop sharply.
Current mainstream VLA models typically concatenate multiple RGB images directly or input them to the policy network in a fixed order, implicitly assuming that the number of cameras, mounting positions, and intrinsics and extrinsics at training and test time are completely identical.
The paper's validation was completed on constructed diagnostic environments and synthetic data, while validation on real physical robots and VLA (Vision-Language-Action) models remains to be done.
All Facts (4)
VLA directly connects pre-trained vision-language models with robot action policies, enabling robots to process natural language instructions, camera images, and low-level motion control in a unified manner.
90%UnverifiedVLA models rely excessively on the camera configuration at training time, and once the number or poses of cameras change in the deployment environment, model performance may drop sharply.
50%UnverifiedCurrent mainstream VLA models typically concatenate multiple RGB images directly or input them to the policy network in a fixed order, implicitly assuming that the number of cameras, mounting positions, and intrinsics and extrinsics at training and test time are completely identical.
50%UnverifiedThe paper's validation was completed on constructed diagnostic environments and synthetic data, while validation on real physical robots and VLA (Vision-Language-Action) models remains to be done.
50%