Synopsis: A robot can have a capable AI model and still lack the knowledge needed to work safely on a factory floor. Equipment relationships, maintenance history and the difference between a documented procedure and actual practice all matter. Physical AI needs that operational context to turn general reasoning into useful action. Keeping the context current is as important as choosing the model that uses it.
Mike Vizard talks with Sudhanshu Gaur, senior vice president and deputy general manager of research and development at Hitachi America, about how to build that understanding into industrial AI. Gaur describes a context layer that combines documentation, live operational data and knowledge held by experienced workers. Agents can examine the available information, assess their confidence and bring in subject-matter experts where gaps remain. Rather than maintaining a separate specialized model for every setting, he favors pairing improving general-purpose models with knowledge that reflects the operation as it exists today.
That context also becomes something worth protecting. An attacker who alters it could steer an agent toward an unsafe decision; someone who extracts it could gain a detailed picture of critical infrastructure. Gaur calls for IT and operational technology teams to work together on the architecture and its safeguards. He also separates flexible reasoning from physical execution. AI can interpret conditions and flag potential problems while established controls, safety interlocks and deterministic workflows continue to govern what equipment actually does.
Simulation gives teams a way to test those decisions before they reach real machinery. Gaur describes using world models to rehearse scenarios, with orchestration and traceability connecting the reasoning, context and proposed action. Robots and drones need to operate within that shared system rather than become isolated deployments. People remain involved both in maintaining its operational knowledge and in making sensitive decisions. Automating dangerous or repetitive tasks does not remove the need to understand what happens when a decision affects workers, equipment or essential services.
