Physics-based AI reasons about how matter behaves. It understands force, motion, constraints, process parameters, and tolerances. Instead of replaying a scripted motion or predicting an outcome statistically, it generates actions based on physical laws.
Physics-based GenAI for Robot Programming
Q&A with Massimiliano Moruzzi, CEO | Xaba
Can you share your background and what led you to found Xaba?
My background is in aerospace engineering and advanced manufacturing. Early in my career, I worked on composite fiber placement systems for the Boeing 787 Dreamliner, where precision, safety, and reliability were non-negotiable. I later led digital manufacturing initiatives for companies such as Ferrari, Airbus, and Lockheed Martin, which provided me with firsthand experience of how automation operates on the factory floor.
Across these environments, I consistently encountered the same issue. Robots were mechanically capable but cognitively fragile. Any small variation in materials, tooling, or setup required weeks of reprogramming. That gap between physical capability and real intelligence is what led me to found Xaba. The goal was to equip machines with a ‘synthetic brain’, providing robots with reasoning grounded in physics, rather than just adding more manual programming code.
What problems did you see in industrial automation that made you believe a new approach was needed?
Industrial automation has been built on the assumption that the environment must be perfectly controlled. Any variation in material, tooling, or setup often causes systems to fail. The standard response has been to add more expensive fixtures, more code, more rules, and more calibration.
That approach does not scale. It creates long deployment cycles, high engineering costs, and fragile systems. The real issue is not a lack of instructions, but a lack of reasoning. Robots are executing logic without understanding physical cause and effect, which is why a fundamentally different approach was needed.
What is physics-based GenAI, and how does it differ from traditional robot programming and other AI approaches used in manufacturing?
Traditional robot programming relies on predefined paths, process parameters, and sequences of actions. It assumes the world will behave exactly the same every time, like in a CAD model. Most AI systems used today focus on pattern recognition and prediction, which works well in digital domains but not in physical ones.
Physics-based AI reasons about how matter behaves. It understands force, motion, constraints, process parameters, and tolerances. Instead of replaying a scripted motion or predicting an outcome statistically, it generates actions based on physical laws. This allows machines to act with intent and adapt when conditions change.
How does Xaba allow robots to generate and validate their own process logic for tasks like welding, drilling, and machining?
Xaba replaces hand-written robot code with a physics-based reasoning system.
Instead of programming every motion, engineers define the desired outcome—such as a weld quality, hole drilling tolerance, cable connection quality, electronics assembly, or material removal—and Xaba’s system generates the process logic automatically.
The robot simulates the task using real-world physics, validates it against constraints like force, temperature, pressure, voltage, and material behavior, and then continuously adapts in production as conditions change. That’s what allows machines to execute complex operations like Server & Data Center assembly, welding, drilling, and machining reliably—without the need for weeks of manual programming or constant recalibration.
Xaba recently opened a new lab in San Francisco. What capabilities does this lab enable, and how does it support the development and deployment of Xaba’s technology?
The San Francisco lab is specifically focused on robotic automation for server and data center fabrication and assembly, enabling us to work directly with real production equipment rather than relying solely on simulations. We validate our Physics-AI on actual hardware across critical processes such as precision assembly, handling, and integration of data center components under realistic, production-scale conditions. This is essential to ensure that our technology performs reliably in the environments where it will be deployed.
The lab also serves as a hub for collaboration with leading partners and manufacturers interested in advancing automation for data center production. By demonstrating our technology on real robotic systems, we significantly shorten the gap between development and deployment, ensuring that what we build is ready for immediate adoption in next-generation data center manufacturing lines
Many automation systems struggle when real-world conditions change, such as material variation or tool wear. How does Xaba’s system adapt in real time without manual reprogramming?
Xaba monitors and reasons about physical conditions as the task is being executed, not after a failure occurs.
The system continuously evaluates forces, geometry, material response, and tool condition during operation. When it detects a deviation—such as material inconsistency or tool wear—it revalidates the process logic in real-time and adjusts the execution parameters accordingly.
For example, if a welding torch begins to wear or a material batch behaves differently mid-shift, the robot compensates automatically without stopping the line.
Because adaptation occurs at the physical level, manufacturers don’t need to pause production, retrain staff, or rewrite code. The robot maintains quality and repeatability even as conditions drift, which is essential for real-world factory environments.
Manufacturers are dealing with labor shortages, reshoring, and higher expectations for quality and repeatability. How does Xaba’s approach help address these challenges?
Xaba reduces dependence on scarce automation experts by eliminating much of the manual programming and calibration work. This allows manufacturers to deploy automation more quickly, even with smaller engineering teams.
At the same time, physics-based reasoning improves quality and repeatability because robots respond to real conditions, not assumptions. This makes it easier to reshore production, maintain consistent output, and scale operations without increasing complexity. Ultimately, it enables human expertise to move upstream toward design, supervision, and innovation, rather than being constantly consumed by troubleshooting and maintenance.
The content & opinions in this article are the author’s and do not necessarily represent the views of ManufacturingTomorrow
Featured Product
