Industrial AI Podcast Why do we need a digital cousin for AI and Robotics?
Mar 4, 2026
Leonardo Barcellona, a robotics researcher at the University of Amsterdam focused on reward and world models, discusses the 'digital cousin' idea as a more varied, robust alternative to perfect digital twins. He explains how using diverse simulated relatives improves real-world transfer. Short, concrete examples and implications for training, reconstruction, and industrial deployment are explored.
AI Snips
Chapters
Transcript
Episode notes
Retrain Workers For Tool-Driven Automation
- Train people to work with higher-level AI tools rather than only teaching low-level PLC or ladder programming.
- Peter suggests shifting training from 100 hours of ladder to shorter introductions plus tool-focused learning with agents.
Wheels Beat Legs For Industrial Humanoids
- Wheeled humanoid robots are a practical industrial compromise because biped locomotion is costly and offers little advantage on factory floors.
- BMW trials Hexagon Robotics' rolling humanoids to handle repetitive or unpleasant tasks in Leipzig.
Agents Shift Roles From Tool To Actor
- Agents blur the line between tool and actor by taking over tasks once done by humans, shifting roles rather than simply augmenting them.
- Peter contrasts spreadsheets (tools) with agents that can perform full analyses previously done by human experts.