Ant Group AI Unit Claims Breakthrough in Robot Sensing with Edge-Focused Vision Model

The embodied artificial intelligence unit of Ant Group has announced a new perception model for robots that it claims outperforms a dominant rival from Meta Platforms while using significantly fewer resources. According to a research paper published by the team behind its Robbyant platform, the LingBot-Vision model surpassed Meta’s 7-billion-parameter DINOv3 across multiple metrics on the NYUv2 depth-estimation benchmark.

The Ant Group team says its model accomplished this feat using just one-seventh as many parameters and less than a third of the training data compared with the Meta system. The achievement points to a more efficient approach to training vision models for three-dimensional spatial understanding.

Precision Edge Detection Key to Advance

What sets the new model apart, according to the firm, is that it was the first of its kind trained specifically to recognize the edges of objects. This specialized focus allows the AI to pinpoint boundaries with high precision—down to a fraction of a single pixel—giving robots a sharper and more accurate understanding of the 3D spaces around them.

The precise edge detection is critical for tasks such as distinguishing transparent surfaces like glass from their surroundings, potentially reducing incidents of robotic collisions and crashes. The company suggests this breakthrough could have practical applications in logistics, manufacturing, and service robotics.

Embodied AI Progress for Robotics

The new vision model serves as the core engine driving an updated depth-perception system called LingBot-Depth 2.0. The development marks a step forward in Ant Group’s broader push into embodied AI, where AI systems are integrated into physical robots that interact with the real world. The firm states that more efficient and accurate perception models are essential for making robots safer and more capable in everyday environments.

The source for this article is https://www.scmp.com/tech/big-tech/article/3359747/glass-crashes-slashed-ant-group-embodied-ai-unit-claims-breakthrough-robot-sensing.