ROYAL Talk&Lunch - Çetin Meriçli

Dear all,

 

You're cordially invited to the following exciting ROYAL seminar by Dr. Çetin Meriçli:

 

Location: South Campus, Faculty of Engineering / 5th Floor,  Hall VYKM 2 Abstract: Physical AI, robots that perceive, decide, and act in the real world, often looks tractable in demos, but becomes brutally difficult in commercial deployments where efficiency, ROI, safety, and robustness are what matters most, not individual capabilities. Warehouses are a perfect stress test: uneven floors, imperfect pallets, variable lighting, human traffic, strict safety requirements, and a long tail of exceptions that happen daily. In this talk, I’ll share the hardest problems we encountered while building and deploying Atlas Robotics’ first Physical AI product, our AI-operated pallet-jack LeVO, and the approaches we developed to solve them in production.

A core theme is problem obsession over solution attachment. In robotics, it’s easy to fall in love with a method: an algorithm, a model architecture, a sensor, and end up optimizing what’s convenient or intellectually stimulating rather than what the job truly demands. At Atlas, our north star is simple: safely replace human work in real warehouses at scale. That framing forces clarity on what matters, exposes the failure modes that dominate ROI, and reveals what problem we must solve next, not what we want to solve.

I’ll cover how we approached problem/market/customer identification, deployment-driven autonomy challenges (perception, localization/navigation, safety), how we handle exceptions without turning autonomy into teleoperation, and how the real world experiences while performing a task as well as tele-assist interactions can become a learning loop where every intervention could potentially be a training signal. I’ll close with practical lessons from building R&D and companies in both the U.S. and Turkey, and what it takes to turn research-grade ideas into a commercial Physical AI platform.