双重突破：全球首个零售VLA大模型来了！开源OpenWBT让机器人遥操门槛暴降！

Core Viewpoint - The article highlights the significant advancements in embodied intelligence showcased at the 2025 Beijing Zhiyuan Conference, particularly focusing on the capabilities of the Galbot robot and the GroceryVLA model developed by Galaxy General Robotics [1][4][11]. Group 1: Event Overview - The 2025 Beijing Zhiyuan Conference took place on June 6-7, gathering leading research institutions, technology companies, and open-source communities in the field of embodied intelligence [1]. - Dr. Wang He, an assistant professor at Peking University and founder of Galaxy General Robotics, participated in the opening roundtable forum and presented the Galbot robot during the main forum [2][4]. Group 2: Galbot and GroceryVLA Presentation - The Galbot G1 showcased its capabilities by autonomously retrieving beverages from a shelf based on voice commands, demonstrating no remote control or prior scene data collection [5][6]. - The GroceryVLA model, which powers the Galbot, was highlighted for its ability to operate in real commercial environments, showcasing its robust performance in complex scenarios [3][8]. Group 3: GroceryVLA Core Capabilities - Strong Applicability: The GroceryVLA employs an end-to-end model architecture that allows it to autonomously identify and retrieve products from densely packed shelves without the need for path planning [13]. - High Generality: The model can handle various packaging types without needing individual adjustments, supporting a unified strategy for different product shapes [15][17]. - Cross-Scene Generalization: Trained on extensive simulation data, the model can adapt to new environments, maintaining stability under varying conditions [18][21]. - Autonomous Decision-Making: The GroceryVLA can dynamically determine the optimal item to retrieve based on task requirements, showcasing advanced task understanding [23][24]. - Strong Interference Resistance: The model can adjust its strategies in real-time to respond to disturbances, ensuring efficient task completion even in challenging environments [27][28]. Group 4: OpenWBT System Launch - Galaxy General Robotics launched OpenWBT, the world's first fully open-source, multi-model, full-body remote operation system for humanoid robots, aimed at enhancing the deployment of humanoid robots in various scenarios [30][33]. - OpenWBT allows for rapid deployment, requiring only a VR headset and a standard computer, significantly lowering the barriers for users [35][36]. - The system supports multiple robot models and facilitates both real-time control of physical robots and remote operation in virtual environments, promoting efficient data collection and model training [37][38]. Group 5: Future Outlook - Galaxy General Robotics aims to lead the transition of humanoid robots from experimental phases to practical applications, enhancing the integration of embodied intelligence technology across various industries [40][41].