Robbyant, the embodied AI unit of Chinese fintech giant Ant Group, has introduced advanced vision technology that boosts robotic perception accuracy, overcoming difficulties with transparent and reflective surfaces.
- New models improve robot accuracy on glass and mirrors
- LingBot-Depth 2.0 tops 12 out of 16 global depth benchmarks
- LingBot-Vision outperforms larger competitor with fewer resources
What happened
Robbyant, a division of Ant Group focused on embodied artificial intelligence, announced two new vision models: LingBot-Depth 2.0 and LingBot-Vision. These models are designed to enhance robots’ ability to perceive and map real-world environments accurately, including traditionally difficult transparent and reflective surfaces such as glass and mirrors. The next-generation spatial perception model LingBot-Depth 2.0 was trained on 150 million samples and demonstrated leading performance on a variety of global benchmarks.
LingBot-Vision is a foundational visual model that powers LingBot-Depth 2.0, distinguishing itself through structural efficiency. Unlike larger models like Meta’s DINOv3, LingBot-Vision achieves better metrics on depth estimation benchmarks while utilizing significantly fewer parameters and training data. Robbyant has also open-sourced LingBot-Vision’s model weights to foster industry collaboration and support commercialization efforts in robotics AI.
Why it matters
Robotic perception of transparent and reflective surfaces has long been a major challenge in the field of embodied AI. Traditional sensors and cameras often fail to accurately gauge distances to these surfaces, leading to frequent miscalculations and operational errors. By addressing this bottleneck, Ant Group’s new models enable robots to interpret 3D environments more reliably and interact safely with complex real-world objects.
This advancement could accelerate the adoption of robots in industries requiring precise spatial awareness, such as manufacturing, logistics, and service robotics. Moreover, the breakthrough in efficient model design shows a path forward for developing powerful AI systems without massive computational costs, making cutting-edge robotics technology more accessible and scalable.
What to watch next
Monitor how Robbyant’s models influence commercial robotics deployments across China and potentially globally, especially in sectors where robotic navigation around transparent materials is critical. The open-sourcing of LingBot-Vision’s weights invites collaboration that could speed innovation and integration with existing robotics platforms.
Keep an eye on competitive developments from other AI labs, including those supporting larger foundation models like Meta’s DINOv3, as efforts continue to balance scale, efficiency, and accuracy in embodied AI. Ant Group’s continuing investment in robotics foundation models since its launch in late 2024 signals a strong commitment to capturing market opportunities in the fast-growing robotics and AI sector.