Les Karpas on why robotics is still waiting for its ChatGPT moment

The primary hurdle for physical AI is a lack of universal data. While OpenAI and Anthropic leveraged the vast expanse of the internet to train language models, robotics lacks an equivalent training set. Even advanced self-driving fleets, such as those operated by Waymo, have spent years accumulating road-specific data that remains difficult to scale across different environments and hardware forms.

Les Karpas, as Nvidia Inception’s Global Head of Physical AI, sits at the center of this challenge. His role involves coordinating with startups across automotive, manufacturing, and smart city sectors that are now racing to bridge this gap through simulation, synthetic data, and foundation models. At the Moscone West event, Karpas will be joined by leadership from companies including Shield AI, Colossal Biosciences, FieldAI, and Foxglove to discuss how to overcome the current hardware-software divide.

Karpas brings a diverse background to this technical bottleneck, having held positions ranging from manufacturing engineer at Stanley Black & Decker to startup CEO and corporate venture capitalist. His session will explore the cross-disciplinary skills required to move robotics from the lab into the real world, offering a look at how Nvidia envisions the next phase of physical AI development.