Script Planner Assistant▸ 
AI Apps 
AI News 
GO-1: China's First General Embodied Foundation Model by AgiBot
GO-1: China's First General Embodied Foundation Model by AgiBotIntroduction
What is GO-1?
Genie Operator-1 (GO-1) is the Zhiyuan Qiyuan foundation model, China's first general-purpose embodied base model, released by AgiBot (Zhiyuan Robotics) on March 10, 2025. The GO-1 model pioneered the Vision-Language-Latent-Action (ViLLA) architecture, composed of a multimodal large model (VLM) and a Mixture of Experts (MoE).
Technical Architecture
ViLLA Architecture: The VLM leverages massive internet image-text data to acquire general scene perception and language understanding capabilities; the Latent Planner within the MoE gains general action understanding capabilities through cross-embodiment and human operation video data; the Action Expert achieves precise action execution through millions of real-robot data points.
Human Video Learning: GO-1 can learn by combining internet videos with real human demonstrations, enhancing its understanding of human behavior.
Few-Shot Rapid Generalization: Capable of quickly adapting to new scenarios and tasks with minimal data or even zero-shot, lowering the barrier to embodied intelligence.
Cross-Embodiment Application: Supports deployment to different robot embodiments, enabling rapid adaptation and efficient implementation.
Continuous Evolution: Continuously learns and evolves from real-world execution through a data feedback system.
Key Features
Human Video Learning: GO-1 can learn by combining internet videos with real human demonstrations, enhancing its understanding of human behavior to better serve people.
Few-Shot Rapid Generalization: Possesses powerful generalization capabilities, able to generalize to new scenarios and tasks with minimal data or even zero-shot, reducing the usage threshold and post-training cost of embodied models.
One Brain, Multiple Forms: As a general-purpose robot policy model, GO-1 can transfer across different robot forms, quickly adapting to different embodiments and achieving collective intelligence upgrade.
Continuous Evolution: Combined with AgiBot's data feedback system, GO-1 can continuously learn and evolve from problem data encountered during real execution, becoming smarter the more it is used.
Experimental Performance
Tested on five tasks of varying complexity, GO-1 achieved an average success rate 32% higher than the existing state-of-the-art models. It performed particularly well on the "Pour Water," "Table Bussing," and "Restock Beverage" tasks.
Application Prospects
Home Scenarios
Household Chores: GO-1 can perform household cleaning, item organization, kitchen tasks, etc., such as pouring water, toasting bread, and tidying rooms.
Intelligent Assistant: Through natural language interaction, helps family members complete daily tasks such as fetching items and turning appliances on/off.
Commercial Scenarios
Retail Services: In retail stores, GO-1 can provide personalized services based on customer needs, such as shelving items, inventory counting, and customer guidance.
Food Service: In restaurants, robots can complete tasks such as serving food, clearing tables, and restocking beverages.
Industrial Scenarios
Logistics and Sorting: Simulating industrial warehouse and production line environments, performing material sorting, packaging, logistics handling, and other tasks.
Production Assistance: Assists on production lines with repetitive or precise operational tasks such as assembly and inspection.
Office Scenarios
Document Management: Helps organize documents and deliver items, improving office efficiency.
Meeting Assistance: Assists in meetings by fetching items and recording information.
Other Potential Scenarios
Education and Training: As a teaching tool, helps students understand robot operations and programming.
Medical Assistance: Assists in hospitals or rehabilitation centers with simple tasks such as transporting items and aiding rehabilitation training.
Zhiyuan Qiyuan Foundation Model GO-1 Project Official Website: https://agibot-world.com/blog/go1
GO-1 Paper: https://agibot-world.com/blog/agibot_go1.pdf
