Robot Dog AI Inspection
A Unitree Go2 robot dog navigates autonomously to a piece of failed equipment, films it, and a multimodal AI diagnoses the fault from video and audio, then auto-creates a purchase requisition in SAP S/4HANA.
Built for the SAP UK Experience Centre to demonstrate what autonomous asset inspection looks like when AI, robotics, and ERP are genuinely connected.
The scenario: a pump has failed in an area that’s either hazardous or unavailable for immediate manual inspection. The robot dog is dispatched autonomously, navigates to the asset, and streams its camera feed back to a web interface. That feed is sent to a multimodal AI model via SAP AI Core for analysis. The first pass identifies an obvious visible fault, establishing the concept cleanly. The second pass goes further: audio from the robot is included, and the AI detects high-frequency bearing noise indicative of advanced wear in the main bearing assembly. It recommends a specific replacement part. That recommendation then triggers the creation of a purchase requisition directly in SAP S/4HANA. Diagnosis to procurement, with no human in the loop.
The two-beat structure is deliberate. The visible fault is accessible; anyone in the room understands it immediately. The audio diagnosis is the moment that lands, because it’s doing what an experienced technician would do: not just looking at the asset, but listening to it. A multimodal model doing real diagnostic reasoning, not pattern-matching on a still frame.
The web interface runs on SAP BTP and connects to the robot over WebRTC, with the full control and inspection surface available in a browser: live camera feed, LiDAR point cloud, manual controls, and the Joule-branded AI assistant handling all the diagnostic interaction.