Robbyant open-sources LingBot-Vision for embodied AI

LingBot-Vision demonstrates that targeted pretraining objectives like masked boundary modeling can yield parameter-efficient visual foundations for robotics. In terms of parametric efficiency, the 0.3B ViT-L model matches the spatial performance of Meta's DINOv3-7B, proving that larger models are not always necessary for fine-grained depth estimation. Furthermore, releasing the weights and source code under the Apache-2.0 license ensures open-source accessibility for academic and commercial integration. This release ultimately provides a robust vision backbone to support more complex physical AI and vision-language-action controllers.