Researchers in the UK have developed a new AI-based training method that helps robots transfer skills learned in simulation into real-world environments with far less physical testing. The approach could reduce the cost and risk of training robots for industrial tasks such as battery recycling, manufacturing, and hazardous operations.

The system was co-developed by researchers at Aston University and the University of Birmingham to address a long-standing robotics problem known as the “sim-to-real gap.” The issue occurs when robots trained in virtual simulations fail to perform reliably in real-world conditions because physical environments contain unpredictable variables such as sensor noise, changing materials, or unexpected forces.

To solve the problem, the researchers used AI to generate variations in environmental conditions during robot training. This allows robots to adapt more effectively once deployed outside simulation.

The team said the method combines the speed and efficiency of simulation with a smaller amount of real-world training data, reducing the need for expensive or potentially dangerous testing.

Closing simulation gaps

Instead of relying entirely on physical training, robots can now learn complex tasks virtually before refining those skills in real environments. The researchers demonstrated the approach using manipulation and cutting tasks that involve physical interaction with materials.

According to the team, the system enables more stable and adaptive robot behavior without requiring massive amounts of real-world data collection.

“By using AI to generate variations in conditions, the new training technique allows robots to transfer skills learned in simulation into the real world much more reliably, using only a small amount of real-world data,” the researchers explained in the release.

The technology could prove particularly useful in industries where direct testing is difficult or unsafe. One example highlighted by the researchers is lithium battery recycling, where robots may need to operate around damaged or hazardous battery cells.

The work was supported through the REBELION project, funded by UK Research and Innovation as part of a wider European effort focused on automated and safe lithium battery recycling systems.

Smarter industrial robotics

Researchers believe the system could eventually help create plug-and-play industrial robots that require minimal reconfiguration before deployment.

“This work shows that we can move beyond purely simulation-based training and achieve reliable performance in real-world conditions with minimal additional data,” said Dr. Alireza Rastegarpanah, assistant professor in applied AI and robotics at Aston University.

“Our long-term vision is to enable plug-and-play intelligent robotic systems that can be trained in simulation and rapidly deployed in new environments with minimal reconfiguration,” he added.

The researchers said the approach could accelerate robotics development across advanced manufacturing, recycling systems, and autonomous industrial operations. Reducing dependence on extensive physical testing may also shorten development timelines and lower operational costs for companies deploying robotic systems.

The team now plans to further refine the method for broader industrial applications where robots must operate under uncertain or changing conditions.

The research was published in Scientific Reports.