AI Technology: Robotic process automation; Generative Language Modeling
The project team is investigating a new model paradigm to enhance machines’ ability to understand and interact with 3D space. The study introduces a spatial language that rethinks 3D representation from a linguistic perspective. Instead of encoding geometry through hidden vectors, the team expresses a 3D scene as a structured sentence whose symbols directly correspond to spatial elements.
This formulation provides an interpretable and compositional view of 3D space, allowing a language model to learn spatial reasoning using the same mechanisms that support linguistic reasoning.
This study has corresponding preliminary experimental results, and the team plans to scale it up and apply it in real-world environments.