The CITIC Securities Research Report points out that the general model is being extended from high-level task planning for robots to move generation and execution control. The robot brain mainly relies on special models such as VLA and WAM that have been trained with robot motion data to transform visual and language instructions into actions. Recently, research institutes such as NVIDIA and Stanford have successively explored the application of GPT-6 Astra to robot control, trying to use the visual understanding, spatial reasoning, and task planning capabilities of general models to reduce dependence on specialized strategy training. Astra has shown good generality in the data set for embodied tasks. We believe that looking forward to 2027, the entry of general models into the field of embodied tasks will become the next investment direction for the AI industry, and general models and dedicated models are more likely to form a division of labor and collaboration in the short term.

The CITIC Securities Research Report points out that the general model is being extended from high-level task planning for robots to move generation and execution control. The robot brain mainly relies on special models such as VLA and WAM that have been trained with robot motion data to transform visual and language instructions into actions. Recently, research institutes such as NVIDIA and Stanford have successively explored the application of GPT-6 Astra to robot control, trying to use the visual understanding, spatial reasoning, and task planning capabilities of general models to reduce dependence on specialized strategy training. Astra has shown good generality in the data set for embodied tasks. We believe that looking forward to 2027, the entry of general models into the field of embodied tasks will become the next investment direction for the AI industry, and general models and dedicated models are more likely to form a division of labor and collaboration in the short term.