GTC Berlin Keynote Highlights Next-Gen Robotics

NVIDIA Chief Executive Officer Jensen Huang brought an unconventional element to his keynote address at GTC Berlin, introducing a quartet of small, autonomous walking robots dubbed "Microducks" to the stage.

The demonstration, which took place on Wednesday evening, highlighted advancements in physical AI development. The four miniature robots autonomously walked onto the stage and assembled before the CEO, sequentially executing commands to sit, stand, and lie down in a synchronized display.

Built through a collaboration between AI community platform Hugging Face and France's Pollen Robotics, each Microduck stands approximately 10 inches tall and weighs less than two pounds. The demonstration offered attendees a live look at how artificial intelligence models can be deployed to govern physical hardware.

Training via Simulation and Cloud Infrastructure

Following the live interaction, NVIDIA detailed the technological pipeline behind the Microduck project. The robots were developed utilizing NVIDIA's Isaac Lab and Isaac Sim environments, alongside the Newton physics engine.

Prior to the stage appearance, NVIDIA demonstrated how an AI agent designed a customized training curriculum, testing the virtual iterations within Isaac Sim before scaling the workload across cloud-based supercomputers. The training outcomes were subsequently verified on an RTX Spark mobile workstation before being deployed to the physical hardware.

"These systems were created on RTX Spark, trained in the cloud using NVIDIA supercomputers, and taught their capabilities leveraging digital twins," Huang explained during the presentation.

Open-Source Architecture for Developers

Designed as an open-source initiative, the Microduck platform aims to provide students, researchers, and developers with an accessible framework for teaching physical skills to machines. The project serves as a tangible benchmark for transitioning AI agents from simulation models into real-world operational robotics.