When machines such as AGVs (automated guided vehicles) work alongside humans in factories and warehouses, they often take precedence and this can impair efficiency due to poor coordination and deadlock situations, Mitsubishi argues. The company’s “cooperative AI” technology is aimed at improving efficiency in such environments.
For example, it allows AGVs to use images from video recordings of these areas to learn and imitate the actions of humans in the area. By learning actions such as yielding, the technology will help the AGVs to avoid unwanted situations such as collisions or stalemates.
To enable AI to learn and imitate human actions, conventional machine learning requires large amounts of operational data – in this case, video data – which takes time and is costly. The IRL technology enhances the process, reducing the amount of video data needed. In simulations, Mitsubishi found that IRL needed 10% or less of the video data normally required.
An example of how Mitsubishi's cooperative AI technology could help to improve machine-human interactions in industrial plants
The in-house simulations are said to have improved operational efficiency by about 30% compared to conventional mixed-work environments populated with less intelligent machines.
Mitsubishi plans to refine the AI technology by testing it on robots and AGVs at production and distribution sites where these machines operate alongside humans. It predicts that the technology will help workers to maintain social distancing, as well as allowing machines and humans to operate safely alongside each other on production lines and in logistics warehouses, as well as in autonomous vehicles.
