What if robots could learn jobs in virtual worlds first — and step into real factories already prepared with less testing?
Researchers in the UK have developed an AI-based robot training method that helps machines transfer skills learned in simulations to real-world environments with less physical testing. The approach could reduce costs, improve safety, and speed up deployment for industrial robots used in manufacturing, recycling, and hazardous operations.
The system, created by researchers at Aston University and University of Birmingham, is designed to solve the “sim-to-real gap” in robotics. Robots trained in virtual environments often struggle once deployed in the physical world, where variables such as sensor noise, material differences, and physical forces can disrupt performance.
To address the problem, the researchers used AI to generate a range of environmental variations during simulation training. By exposing robots to changing conditions before deployment, the system improved the machines’ ability to adapt and perform in real-world settings.
The technique combines simulation learning with a smaller amount of real-world data, reducing the need for physical testing. Instead of relying entirely on physical training, robots can first learn tasks in simulations and then refine their performance through limited real-world interaction.
Researchers tested the method on robotic manipulation and cutting tasks that required contact with physical materials. They said the approach produced more reliable robot behavior while reducing the amount of real-world data collection required.
The technology could be useful in industries where testing conditions are risky or difficult to replicate. One example is lithium battery recycling, where robots may need to handle damaged or hazardous battery cells.
Researchers said the system could support plug-and-play industrial robots that require little reconfiguration before use. The team expects the approach to accelerate development in manufacturing, recycling, and autonomous industrial operations while reducing development timelines and operational costs.
Future work will focus on adapting the technology for industrial environments where robots must operate under changing conditions.
