Recently, Galaxy General Technology Company, in collaboration with Beijing Zhiyuan Artificial Intelligence Research Institute, Peking University, and the University of Hong Kong, has officially launched an end-to-end embodied grasping foundation model named GraspVLA. The introduction of this model marks a significant breakthrough in embodied intelligence technology, showcasing comprehensive capabilities in perception, learning, and interaction with the environment.
The training process of GraspVLA consists of two phases: pre-training and post-training. The pre-training phase is based on a billion frames of data, ensuring the model's high generalization ability and zero-shot testing capability across various scenarios. This means that the model can demonstrate excellent grasping performance even in unseen environments and tasks. The enhancement of this capability indicates that the flexibility and adaptability of intelligent robots in real-world applications will be greatly improved in the future.
Image Source Note: Image generated by AI, authorized by service provider Midjourney
The officials also released seven generalization "gold standards," providing a new evaluation basis for the industry. These standards not only offer clear goals for developers but also guide researchers in model optimization and practical applications. Galaxy General stated that the launch of GraspVLA will provide strong support for robots' autonomous operations, object recognition, and interactions in complex environments.
With the continuous advancement of artificial intelligence technology, the application prospects of embodied grasping technology are vast. As a pioneer in this field, GraspVLA will play an important role in various industries such as warehousing logistics, manufacturing, and medical services. In the future, robots will no longer be limited to simple repetitive tasks; they will be able to learn and adapt to different environments, complete more complex tasks, and even engage in more natural interactions with humans.
At the press conference, participating organizations also expressed their commitment to continue promoting the development of this field and exploring more practical application scenarios. The release of GraspVLA symbolizes a solid step forward for intelligent robots into a new era of "understanding grasping."
