therobotreport.com
Nov 11, 2026
Can AI agents program industrial robots without training data? Berkeley says yes
AI-summarised brief · reviewed before publication
UC Berkeley and NVIDIA unveiled RoboSimStudio, an open‑source platform that lets AI agents generate, test, and refine industrial robot programs without extensive training data. The system uses large language models to write code, while modular agentic frameworks—GaP and VABAR—break tasks into perception, grasping, and motion skills and benchmark them across six industrial scenarios. Demonstrated at CoRL 2026, the tool produced a realistic Newton physics simulation from a video in under an hour, showcasing rapid, iterative development of robot behavior.
💡 Why It Matters
- · By automating code creation and simulation, RoboSimStudio could slash the weeks‑to‑months cycle that currently delays robot deployment, enabling faster, more reliable production line integration.
