For many years automation has played a vital part in the manufacturing industry. From the moment automated technology was introduced it’s sped up production while costing less. Now BMW is teaming with MIT researchers to find an even more efficient way of implementing automated systems into the production line – with the help of new AI tech.
The purpose of the project is to find a better way of allowing both robots and humans to work together in the most efficient manner possible.
Such experiments originally began in 2018, with the team rigging up a robot capable of delivering car parts across the factory using a rail system. Since human workers also occupy the floor, the robots would need to be aware of their surroundings in order to prevent any accidents which could happen in classic automated factories.
The original technology used a sensor system that would follow human workers across the factory floor, and then pause once they walked within proximity of the robot. A good idea, but unfortunately not the most time efficient, with the robot often taking unnecessary pauses.
This led to the teams current efforts – developing highly accurate prediction algorithms capable of tracking humans on the factory floor by predicting their movement in real time.
Researchers on the team say that current algorithms are simply not accurate enough to predict proper human movement, with most using a dot method which marks out points on the persons path. Due to the messy nature of human movement, these algorithms are often confused when, say, someone walking decides to stop for a moment before continuing.
Instead of using this, the new algorithm looks at the subjects trajectory and bases predictions off of a library of reference trajectories. This method was then tested in two scenarios; one in which a person crossed in front of a robots path, and one in which a hand would reach across a table and install a bolt, which a robot would then install.
From these tests, the team found that the robots were able to recognise the correct timing of the subjects walking around, and rather than pausing for prolonged times, would start working very shortly after the subject had walked past it on the factory floor – knowing that the subjects trajectory wasn’t likely to spin back around into the machine.
Julie Shah – associate professor of aeronautics and astronautics at MIT – told MIT News that this type of algorithm will prove useful in the future. Allowing robots to properly predict human behaviour will allow them to interact more efficiently, from factory settings to the home.
