# Glass crashes slashed? Ant Group embodied AI unit claims breakthrough in robot sensing > Source: > Published: 2026-07-07 11:30:31+00:00 # Glass crashes slashed? Ant Group embodied AI unit claims breakthrough in robot sensing New perception model for embodied AI surpasses dominant rival from Meta Platforms at smaller scale with less training, firm claims [Ann Cao](/author/ann-cao)in Shanghai [artificial intelligence](https://www.scmp.com/topics/artificial-intelligence?module=inline&pgtype=article)arm of Chinese fintech giant Ant Group, launched a new vision model that it claims can help robots overcome a long-standing challenge: accurately perceiving glass, mirrors and transparent objects. [Ant Group](https://www.scmp.com/topics/ant-group?module=inline&pgtype=article)on Tuesday unveiled its next-generation spatial perception model, LingBot-Depth 2.0, alongside a new foundational visual model called LingBot-Vision, as AI labs race to equip machines with the “brains” required to navigate complex physical spaces. [robotics](https://www.scmp.com/topics/robotics?module=inline&pgtype=article)by ensuring machines could “accurately and stably” see in unpredictable, real-world environments, according to the Robbyant, which is also known as Ant Lingbo Technology. It was the first model of its kind trained specifically to recognise the edges of objects, the firm said. This allowed the AI to pinpoint boundaries with high precision – down to a fraction of a single pixel – giving robots a sharper understanding of the 3D spaces around them, it added. [Meta Platforms](https://www.scmp.com/topics/meta-platforms?module=inline&pgtype=article)’ open-source vision model, DINOv3. But where that model relies on massive computational scale, LingBot-Vision aims to outrun rivals through structural efficiency. According to a research paper published by the Robbyant team, LingBot-Vision surpassed the 7-billion-parameter DINOv3 across multiple metrics on the NYUv2 depth-estimation benchmark. It achieved this using one-seventh as many parameters and less than a third of the training data, according to the benchmark. The new vision model serves as the critical engine driving LingBot-Depth 2.0.