How Path Robotics Uses Foundation Models to Transform Real-World Welding Automation

Welcome to Robot Learning in Industry, a Neuracore original series exploring how machine learning is transforming real-world robotics. Each episode breaks down the practical challenges, emerging tools, and engineering breakthroughs shaping how ML is deployed outside the lab in factories, warehouses, hospitals, and beyond.

In this episode, we sit down with Nima Gard, Director of AI at Path Robotics, the company pioneering autonomous welding systems and building one of the first foundation models for manufacturing.

Their conversation offers a rare look at how machine learning can unlock automation in one of industry’s most difficult tasks: welding. With an approach centered on unsupervised learning, massive real-world datasets, and custom-built scanning hardware, Path is pushing the boundaries of what physical AI can achieve on the factory floor.

𝗧𝗵𝗲𝘆 𝗲𝘅𝗽𝗹𝗼𝗿𝗲:

• Why welding is the perfect proving ground for industrial foundation models
• The importance of high-fidelity scanning before, during, and after welds
• How every data point, good or bad, strengthens the model and accelerates improvements
• The technical infrastructure required to train at production scale across a global fleet
• How their methods extend far beyond welding to other core manufacturing workflows

Featuring candid insights from a team building robots that operate in high-variability, real-world industrial environments, this discussion cuts through the hype to reveal what is actually working today in ML-driven automation and what breakthroughs are still needed.

𝗖𝗵𝗮𝗽𝘁𝗲𝗿𝘀:
• 00:00 Intro – Who Is Path Robotics?
• 00:14 Building Foundation Models for Manufacturing
• 00:27 Why Welding Still Isn’t Automated
• 00:34 Removing Human Expertise With Unsupervised Learning
• 00:52 How Path’s Custom Scanners Capture Weld Data
• 01:09 Training on Every Weld From the Fleet
• 01:28 Why Good and Bad Welds Are All Valuable Data
• 01:42 Expanding Beyond Welding: General-Purpose Robotics Tech
• 02:23 Applying Welding Automation Lessons to New Tasks
• 02:56 How Path Robotics Built Its Data Flywheel

Learn more about Neuracore: https://www.neuracore.app/

Learn more about ‪@PathRobotics‬: https://www.path-robotics.com/

Subscribe to Robot Learning in Industry for more conversations with the teams bringing machine learning into the real world.

#robotics #ai #machinelearning #robotlearning #physicalAI #industrialautomation #manufacturingAI #factoryautomation #weldingautomation #PathRobotics #Neuracore #foundationmodels #robotarm #computerVision #MLengineering