Scientists train robots to make independent decisions

Scientists have developed a software that allows industrial artificial intelligence (AI) robots with technical vision to set out and adjust the movement trajectories of their tools in real time without reducing given precision levels.

The team from Far Eastern Federal University (FEFU) in Russia developed and implemented a new principle for smart industrial robots control.

According to it, robots are able to set and adjust the trajectories and regimes (speeds) of tools movement on their own while processing details under uncertain conditions and in a changing working environment.

The new software allowed the team to get around 0.5 milimetre (mm) precision in the operation of robotic tools, including the actions that require additional force application.

However, many high-accuracy operations require precision within the 0.2-0.1 mm range, researchers said.

"The issue lies in the imprecise technology used to manufacture the robots themselves, and it hasn't been resolved anywhere in the world yet," said Professor Vladimir Filaretov from FEFU.

"We've already developed a method to eliminate this defect based on special test movements. It proved to be efficient in models, and right now we are working to implement it in practice.

"If we obtain positive results, it would be a breakthrough in the practical application of robots in general. And if no, we'd continue to work until we have a positive result. Generally, this is a working method," said Filaretov.

Using a technical vision system, a machine forms a virtual image of its workspace, recognises each piece, and determines its exact position.

A robot can also identify deformations in large pieces that occur in the course of their fixation. Based on the virtual image, it determines the trajectories of its working tools.

"It's important to emphasise that the methods, algorithms, and software developed by us are of universal nature. They can be used to control almost any types of robots: industrial robots, underwater devices, unmanned ground vehicles, flying, and many promising agricultural robots.

"They only require minor adjustments that are already included into the software and take into account their specific features," Filaretov said.

"Our developments, including smart VR-based control, maximise on the capabilities of modern technologies and are able to increase the efficiency of technological processes by several times while preserving the quality of the products," said Filaretov.

The new smart control method has already been implemented at the Dalpribor plant (Vladivostok) and is currently being tested and adjusted in view of the recent industrial challenges, researchers said.