Industrial AI Podcast The Welding World Model
6 snips
Oct 8, 2025 Andy, the co-founder and CEO of Path Robotics, shares insights from his groundbreaking work in AI-driven welding robots. He discusses the importance of adaptability in manufacturing and how real-world data significantly enhances AI performance. The conversation reveals how Path's Obsidian model integrates multimodal sensors—using everything from sound to vision—for precision. Andy also highlights the challenges of training robots on diverse materials and the impressive outcomes, like a 98% first-pass yield, demonstrating the future of robotics in industry.
AI Snips
Chapters
Books
Transcript
Episode notes
Agentic Systems Need Data Science Backbone
- Agentic systems still need solid data science teams in the loop for success.
- Gabriel Krummenacher argued agentic projects remain fundamentally data-science projects.
Founding Path Robotics From A PhD
- Andy co-founded Path Robotics during his PhD focused on deep RL for humanoid robots.
- He positioned the company in Columbus to be close to manufacturing customers.
Multimodal World Model Drives Adaptive Welding
- Path builds a multimodal sensor and a large model (Obsidian) that maps perception to welding actions.
- They combine imitation learning, a learned world model, and RL to produce adaptive welding policies.
