CITIC Securities: GPT-6 Astra enters robot control as physical AI hits turning point
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
CITIC Securities said in a research note dated October 26 that general-purpose large models are extending from high-level task planning for robots toward motion generation and execution control. Robot brains currently rely mainly on specialized models such as VLA and WAM trained on robot action data to convert visual and language instructions into actions. Recently, research institutions including NVIDIA and Stanford have explored applying GPT-6 Astra to robot control, seeking to use the general model's existing visual understanding, spatial reasoning and task planning capabilities to reduce dependence on specialized policy training. Astra showed relatively good generalization on embodied-task datasets. CITIC Securities expects that by 2027, general models entering embodied-task domains will become the AI industry's next investment direction, while general and specialized models are more likely in the short term to form a division-of-labor and collaborative arrangement.
Source report
October 26 — According to a research note from CITIC Securities, general-purpose large models are extending from high-level task planning for robots toward motion generation and execution control.
The "robot brain" primarily relies on specialized models such as VLA and WAM trained on robot motion data to translate visual and language instructions into actions. Recently, research institutions including NVIDIA and Stanford have begun exploring the application of GPT-6 Astra to robot control, attempting to leverage the general model's existing capabilities in visual understanding, spatial reasoning, and task planning to reduce dependence on specialized policy training.
Astra has demonstrated strong generalization on embodied task datasets. CITIC Securities believes that looking ahead to 2027, general models entering the embodied task domain will become the next investment direction for the AI industry, while in the short term, general and specialized models are more likely to form a division of labor and collaborative relationship.
Source
domesticEastern