AI-generated conceptual illustration of industrial robot work
How far is a robot that can complete a demonstration movement from reliably taking on tasks on a production line? Shanghai's newly announced intelligent manufacturing action plan provides a set of clear targets for observing this distance.
On October 27, the website of the Shanghai Municipal Commission of Economy and Informatization released the "Shanghai Action Plan for Accelerating the Development of a New Generation of Intelligent Manufacturing," with the document dated October 22. The plan proposes driving 8,000 humanoid robots into factories by 2028, exceeding 2,000 AI application scenarios in manufacturing, and launching more than 300 high-level agents.
These figures are deployment targets for 2028, not the scale of applications already completed.
Beyond robots, models and data must also enter the process
The plan lists robots with autonomous perception and decision-making capabilities, industrial machine tools, and inspection equipment as development directions for physical AI equipment. At the same time, for tasks such as design assistance, adaptive production adjustment, and predictive maintenance, it deploys industrial models and agents, promoting the formation of usable data resources from manufacturing enterprises' process knowledge and expert experience.
AIYXL believes these efforts need to be understood together. Equipment must execute actions, models must identify tasks, and process data must tell the system: under what conditions operations can be performed, which results count as qualified, and who should handle anomalies when they occur.
For example, in a part loading and unloading task, successful grasping is only one part of the process. Changes in incoming material position, reflective part surfaces, fixture offset, or a downstream equipment pause may all require the system to adjust its actions or stop work. Whether it can recognize these changes and resume operation is closer to the actual requirements of a factory than the speed of a single demonstration.
From "Entering the Factory" to Asking "What Tasks Will It Take On"
To evaluate subsequent projects, one can start with specific processes: how many effective operations the robot performs each day, at which stages manual intervention occurs, how long changeover takes, and whether failures affect the entire production line.
Input-output should also include integration, maintenance, training, and downtime costs. For factories with different products and different batch sizes, the same type of equipment may not bring the same benefits.
This plan places equipment, models, data, and application scenarios within the same policy framework. What is worth continuing to track next is which tasks can form stable, replicable implementation experience, and whether enterprises are willing to expand applications after validation.
Source: Original action plan from the Shanghai Municipal Commission of Economy and Informatization