**Ant Group’s Lingbo Technology releases spatial perception model LingBot-Depth 2.0**
The AMW Read
Novelty 2: Lingbo is a new player in the robotics vision segment, and the open-source LingBot-Vision model with superior data efficiency meaningfully updates the baseline. Significance 2: The hardware-software bundling strategy with Orbbec could accelerate embodied AI deployment at segment level, bu
**Ant Group’s Lingbo Technology releases spatial perception model LingBot-Depth 2.0**
On July 7, 2026, Ant Group’s embodied AI subsidiary Lingbo Technology (灵波科技) released LingBot-Depth 2.0, a spatial perception model trained on 150 million data points. The model improves edge clarity, small-object recognition, long-range depth estimation, and robustness in complex scenes such as glass, mirrors, and transparent objects. Lingbo also open-sourced its visual foundation model LingBot-Vision in four sizes (ViT-G/L/B/S), which uses a novel pretraining objective centered on boundary structure. The company has partnered with 3D camera maker Orbbec (奥比中光) to integrate LingBot-Depth into camera SDKs and a planned integrated camera product by year-end. LingBot-Depth 2.0 topped 12 of 16 benchmarks in depth completion and halved depth error in the hardest indoor large-depth-missing scenarios.
**Why it matters within the AI Market Watch substrate**
This release fits the recurring pattern of hyperscaler-backed robotics companies using vertical integration to close the perception-to-action loop. Lingbo, spun out of Ant Group, is playing a structural role in the robotics-vision stack that mirrors how hyperscalers have used distribution moats in foundation models. The open-sourcing of LingBot-Vision—trained on just 160 million images yet outperforming DINOv3 in depth estimation—is a data-efficiency signal that challenges the prevailing assumption that larger pretraining corpora are strictly necessary. The partnership with Orbbec, a Chinese 3D-sensor leader, creates an integrated hardware-software bundle that could lower the barrier for robotics startups to deploy reliable spatial perception, accelerating embodied AI deployment in logistics, service, and manufacturing.
**Grounded expert take**
LingBot-Depth 2.0’s performance on traditionally difficult scenes (glass, mirrors, transparent objects) addresses a well-known physics bottleneck in robot vision that has limited real-world deployment of service robots. By combining a compact, open-weight vision foundation model with a commercial depth-completion layer, Lingbo is effectively commoditizing a capability that was previously locked inside expensive sensor rigs. The Orbbec integration—first as a data-collection accessory, then as an embedded SDK, later as a bundled camera—is a textbook acqui-licensing pipeline: it turns a model into a recurring revenue stream through hardware bundling. If the integrated camera ships at competitive price points, it could compress the capital cycle for robotics startups by eliminating the need to build custom perception stacks.
#robotics #spatialAI #embodiedAI #AntGroup #openvision #robotvision #depthperception
