SoFar introduces a language-grounded orientation framework that enables robots to accurately understand and manipulate objects based on their spatial orientation.

• Addresses a critical gap in Visual Language Models by adding orientation understanding
• Combines geometric reasoning with intuitive language-based representation
• Enhances robots' ability to perform fine-grained manipulation tasks
• Bridges spatial perception and physical interaction capabilities

This research significantly advances robotic engineering by enabling more precise, context-aware object manipulation - a fundamental requirement for next-generation factory automation and manufacturing applications.

SoFar: Language-Grounded Orientation Bridges Spatial Reasoning and Object Manipulation