GR00T N1.7
A general-purpose humanoid robot foundation model released by NVIDIA, built on a VLA (vision-language-action) architecture and intended for general policy learning on humanoid robots.
Timeline (last 90 days)
A study of GR00T N1.7 found that when keywords in instructions are replaced with synonyms, the VLA policy may successfully select the target object yet fail to complete the task.
GR00T N1.7 adopts a "VLM backbone + action expert" architecture, in which the connector is the interface module between the vision-language model and the action generator.
VioLA can execute locomotion commands on a real robot without any task-specific fine-tuning, achieving a 100% success rate, compared with 16.7% for GR00T N1.7 and 0% for Ψ₀.
All Facts (3)
A study of GR00T N1.7 found that when keywords in instructions are replaced with synonyms, the VLA policy may successfully select the target object yet fail to complete the task.
50%UnverifiedGR00T N1.7 adopts a "VLM backbone + action expert" architecture, in which the connector is the interface module between the vision-language model and the action generator.
50%UnverifiedVioLA can execute locomotion commands on a real robot without any task-specific fine-tuning, achieving a 100% success rate, compared with 16.7% for GR00T N1.7 and 0% for Ψ₀.
50%