21/09/2026
Capgemini and Intel have created a physical AI demonstration built around the Tower of Hanoi, showing how a robot can see, understand, decide, act, and observe locally at the edge.
The demo was introduced at Intel’s booth at AI Infra Summit 2026 in Santa Clara, September 15 to 17, 2026. It combines computer vision, conventional software, a vision language model, and a vision language action model to move puzzle pieces between pegs one step at a time.
The system uses different compute resources for different workloads. The CPU manages the application and conventional software, the integrated GPU runs the model responsible for robotic movement, and the NPU supports scene understanding.
The setup also includes a live view of the camera feeds and AI outputs, making it easier to see how each action is produced and to investigate unexpected behavior.
How important is observability when AI controls physical machines?