UK researchers create AI method to close robot simulation-to-real gap

Researchers at Aston University and the University of Birmingham in the United Kingdom have developed an AI‑based training framework that enables robots to transfer skills learned in virtual simulations to real‑world tasks with far less physical testing. The approach tackles the longstanding “sim‑to‑real” gap by using AI‑generated variations of environmental conditions and a variational auto‑encoder to match simulated and real data, allowing robots to adapt to differences in lighting, material properties, friction and forces.

Demonstrations on cutting and manipulation tasks show more reliable performance, and the team highlights applications such as battery‑recycling, advanced manufacturing and hazardous operations. The method, part of the REBELION project funded by UK Research and Innovation, aims to create plug‑and‑play industrial robots that can be deployed quickly with minimal re‑configuration. Results were published in Scientific Reports (2026).