How Path Robotics uses AI to optimize robotic welding

The Robot Report Podcast · 10 Jul 2026 · 1h 16m

AI in practice: one real use of AI described

What this episode covers

Andy Lonsberry, CEO and co-founder of Path Robotics, explains how his company uses AI, specifically reinforcement learning and real-time visual feedback, to make robotic welding adaptive and capable of handling complex, non-repeatable industrial environments. He details the development of 'Obsidian,' a neural simulator that allows robots to weld with high first-pass yields without relying on perfect CAD models, and discusses the launch of 'Rove,' a mobile welding platform built on Boston Dynamics' Spot quadruped for shipbuilding and construction. Lonsberry also outlines Path Robotics' 'Robotics as a Service' business model, which delivers welding outcomes rather than just hardware, and notes the company's recent Series D2 funding and rapid scaling.

AI in practice

Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.

ManufacturingObsidian

Path Robotics uses its AI model Obsidian to generate adaptive welding paths and control robot motion in real-time based on visual feedback, allowing mobile robots to weld large, variable parts like ship hulls without pre-programmed fixed paths.

“We make a neural network that simulates the physics of welding. … inside of that model, we train a reinforcement learning agent … we deploy obsidian, and that model is basically the system at the edge of the of our robots. It's taking in the data on that per part basis and making all the choices around every single welding parameter…”

Andy Lonsberry · Hear it at 41:11

Mentioned

Companies
Path Robotics, Boston Dynamics, Agility Robotics, Tiger Data
Products
Spot, Obsidian, Rove, Unitree G1, Da Vinci, Digit, TimescaleDB

Other shows on the same ground, closest first.

  • The Manufacturing Executive · 25m

    Perspectives on Automation, AI, & the Future w/ Andy Lonsberry

    Andy Lonsberry, Founder and CEO of Path Robotics, explains how the company identified a market pain point in high-mix, low-volume welding and developed a fully autonomous robotic system using machine learning and computer vision. He details the journey from a 24-month consulting exploration to securing $15 million in Series A funding by bringing an autonomous robot to Silicon Valley. Lonsberry also discusses the 'Robots as a Service' (RAS) business model, which eliminates upfront capital expenditure for manufacturers by tying payment to system performance, and highlights the abundance of engineering talent in the Midwest.

  • Manufacturing Happy Hour · 42m

    Cars, Welding Robots, and Midwest Startups with Path Robotic's Sheena Patterson

    Sheena Patterson, VP of Operations at Path Robotics, discusses her career journey from Ford and Tesla to leading automated welding robot operations in the Midwest. She explains how Path Robotics utilizes a 'robots-as-a-service' model to provide autonomous welding solutions, avoiding large capital expenditures for customers while maintaining continuous support. Patterson highlights the importance of people management and passion in manufacturing, contrasting the structured chaos of traditional automotive launches with the mission-driven culture of Silicon Valley startups like Tesla and Zoox. She also details how Drive Capital’s investment enabled Path Robotics to scale production, improve inventory management, and attract diverse talent to Columbus, Ohio.

  • Manufacturing Happy Hour · 1h 05m

    231: Why VCs and Startups Are Betting on Manufacturing and the Midwest Featuring Drive Capital and Path Robotics, Live from Columbus, OH

    Andy Lonsberry, CEO and co-founder of Path Robotics, and Nima Gard, Director of AI, explain how their Columbus-based company uses machine learning and data to automate welding and dexterous assembly, while Nils Root of Drive Capital discusses the firm's thesis on investing in Midwest manufacturing innovation. The guests highlight that Path Robotics achieves 3x growth by treating data as its primary asset, training large foundational neural networks on real-world field data to handle manufacturing edge cases and reduce the sales cycle to 80 days. They argue that Ohio offers a superior talent pool and ecosystem for building world-class technology companies without needing to mimic Silicon Valley, emphasizing the importance of a supportive local culture including legal and real estate partners who understand startup risks.

  • Robot Talk · 31m

    Episode 160: Robotic blacksmiths - Edward Mehr

    Edward Mehr, Co-Founder and CEO of Machina Labs, explains how their RoboCraftsman system uses AI-driven robotics to perform flexible, on-demand metal forming for aerospace, defense, and automotive industries. He details the transition from traditional, rigid factories to portable, software-defined 'data center' style manufacturing that eliminates economies of scale constraints. Mehr discusses the technical challenges of sheet metal springback and how the company bridges the simulation-to-reality gap by combining physics-based models with real-world data capture. He also highlights their work with NASA and the Air Force to rapidly repair legacy systems, and a partnership with Toyota to enable mass-customized vehicle bodies.

  • Manufacturing Happy Hour · 49m

    183: A Case Study in the Adoption of Robotics Featuring Rob Goldiez (CEO of Hirebotics) and Justin Montes (CEO of DeAngelo Marine Exhaust)

    Rob Goldiez, CEO of Hirebotics, and Justin Montes, CEO of DeAngelo Marine Exhaust, discuss the adoption of collaborative robots (cobots) for welding. Montes explains how DeAngelo faced a skilled labor shortage and capacity issues, leading to a quick decision to purchase a Hirebotics cobot after successful on-site testing. Goldiez highlights that Hirebotics provides app-based, turnkey welding solutions designed for welders rather than robotic programmers, emphasizing ease of use and rapid support.

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by The Robot Report Podcast. The episode itself belongs to its publisher.