Robbyant Launches LingBot-Depth 2.0 and LingBot-Vision for Robotic Spatial AI

Robbyant, Ant Group's Physical AI company, has unveiled LingBot-Depth 2.0 and LingBot-Vision, two advanced robot vision models designed to improve robotic depth perception and spatial understanding. The new models enhance how robots interpret complex environments, enabling more accurate navigation, object recognition and manipulation across industrial, logistics and embodied AI applications.

Built on the company's previous depth-completion technology, LingBot-Depth 2.0 was trained using 150 million samples and achieved leading performance in 12 of 16 public depth-completion benchmarks. The model significantly improves depth estimation in challenging environments containing transparent objects, reflective surfaces and large regions of missing depth data, delivering greater perception accuracy than its predecessor.

Robbyant stated:

"LingBot-Depth 2.0 extends robotic spatial perception by improving depth completion, edge reconstruction and scene understanding, providing a stronger foundation for embodied AI systems operating in real-world environments."

Supporting the platform is LingBot-Vision, a visual foundation model trained on 160 million images using boundary structure as its pre-training objective. The model enables high-precision object boundary detection, spatial reasoning and downstream vision tasks, while also tracking object contours across video sequences to improve robotic perception and scene understanding.

The company also announced a strategic collaboration with Orbbec, whose Depth Vision Laboratory has certified LingBot-Depth 2.0 following validation on the Gemini 330 stereo 3D camera platform. The partners are developing a commercial SDK for robotics customers and an integrated intelligent 3D camera scheduled to release later this year.

Robbyant has additionally open-sourced the LingBot-Vision model weights on GitHub and Hugging Face, supporting broader adoption of advanced spatial AI technologies across the robotics ecosystem.