Drives and Controls Magazine
AI machine control algorithm slashes set-up times
Published:  24 May, 2017

Mitsubishi Electric has announced a deep-reinforcement algorithm for artificial-intelligence (AI) based control of machinery that, it says, needs just one-fiftieth of the number of trials needed to set up conventional AI-based controls. The new algorithm will allow smart equipment such as industrial robots and vehicles to use sensors and cameras to learn rapidly about their environments to allow finely-tuned AI-based control.

Conventional AI-based machine-learning systems can take up to half a day to set up and require input from human experts, according to Mitsubishi. These systems need time to process the “huge” amounts of data obtained from cameras and sensors, and to perform trials based on this data.

The company says that its new system is fully automated and takes less than 30 minutes to set up.

When combined with the Compact AI technology that Mitsubishi released in February 2016, the new algorithm requires just 1% of the calculations needed for conventional methods. It will also allow machines with limited processing capabilities to perform deep-reinforcement learning.

Mitsubishi says that its AI-based control algorithm will save time and avoid the need for human experts when setting up smart machines

•  Mitsubishi unveiled the new algorithm at an event in Japan showcasing its latest technologies, where it also announced a new brand covering its AI technologies. The Maisart (Mitsubishi Electric AI creates state-of-the-art in technology) brand is aimed at raising awareness of the company’s AI activities. The global AI market was estimated to be worth around $35bn in 2015, and is growing by about 30% a year.