Most AI demos never reach production. In physical AI the gap is wider, because a model that works on a laptop still has to run on constrained hardware, in a noisy environment, on a power and network budget, with someone accountable when it fails.

This session covers what it actually takes to move robotics and edge AI from prototype to deployment. Nolan breaks down the decisions that determine whether a system ships: where inference runs, edge versus cloud tradeoffs, model optimization for local hardware, sensor data pipelines, data center capacity and power, and the operating model needed to support it all. He uses real examples from AARI's training environments and edge AI work, along with lessons from enterprise cloud deployments.

Attendees will leave with a deployment checklist for physical AI systems, a framework for deciding what to run locally versus in the cloud, and a clear view of the infrastructure and staffing that teams consistently underestimate

Format: Talk, 30–45 min. Level: Intermediate. Audience: engineers, architects, technical leaders.