Physical AI and Robotics at NVIDIA GTC 2026 with Diana Wolf Torres

In this talk

1. 4:59

   Efficiency and AI infrastructure

   Diana and Jonathan note a shift toward measuring computing by tokens produced per second and per watt, as electricity limits data center growth. They discuss NVIDIA’s Vera Rubin chips and the claim that their efficiency could make upgrading worthwhile even for operators with earlier chips not yet installed.

2. 8:57

   Agents as tools for robotics work

   Diana describes experimenting with OpenClaw on a DGX Spark and learning how agents might coordinate specialized work. She imagines agents that know Isaac Sim, Isaac Lab Arena, or Cosmos helping orchestrate Lightwheel’s simulation workflow, while local models can reduce token costs and keep company data on-device.

3. 34:56

   Robots that meet practical needs

   Diana highlights Worker’s tile-laying system for Fireclay Tile, which was struggling to hire skilled workers even at $40 to $50 an hour. Worker uses a pick-and-place arm, trains it with Isaac Sim and synthetic data, and charges for productive hours through a robotics-as-a-service model.

4. 16:55

   Safety, coordination, and trust

   The discussion covers autonomous Caterpillar machinery, the safety benefits of keeping people out of active work zones, and the risks posed by large, powerful robots near crowds. Diana points to Disney’s Olaf as an example of planning robot falls around people and training behavior in simulation; she warns that injuries could undermine public trust in robotics.

5. 43:21

   Specialized robots and a growing ecosystem

   The pair describe humanoid robots working together to serve snacks, open robot platforms that add software to machines such as Unitree’s G1, and household robots designed for specific tasks. They also discuss Amazon’s acquisitions, AgiBot’s reported rise to 10,000 humanoid units, and Asimov’s open-source robot design.