Qualcomm on Physical AI: Autonomous Driving Paved the Way, Dexterity Remains Early

ryanshrout · x · 2026-07-30

Nakul Duggan, who leads automotive and robotics at Qualcomm, shared deep insights into the evolution of physical AI and robotics at Computex.

Core Driver: Robotics has existed as a mechanical discipline for 15 years. The recent qualitative change is that it "got a brain," funded largely by a decade of autonomous driving investments in sensors, compute, and training pipelines.
Stages of Development: Physical AI arrives in two stages. Navigating the real world is largely solved, but dexterity in unstructured environments is still in its infancy.
Cloud vs. Edge: There is no debate here. Real-time and safety-critical requirements mean models train and simulate in the cloud but deploy and run locally.
Industry Focus: This is not just a humanoid story. Assembly line robots need eyes and hands, not legs. Qualcomm's Dragonwing IQ10 Robotics Reference Design is built for this broad utility.
Success Metric: The 2-3 year success metric isn't robots shipped, but enterprise dollars spent automating tasks to address labor shortages.

More from Embodied