MIT develops a better way for robots to predict human movement

People and robots working together has tremendous potential for factory and construction site settings, but robots are also potentially incredibly dangerous to people, especially when they’re large and powerful, which is typically the case for industrial robots.

There are plenty of efforts to make “corobotics” a reality, including production machines like the YuMi produced by German robotics giant ABB . But a new algorithm created by MIT researchers could help make humans and robots working together even safer.

Researchers working with automaker BMW and observing their current product workflow noticed that the robots were overly cautious when it came to watching out for the humans in the plant — they’d lose lots of potentially productive time waiting for people to cross their paths long before there was any chance of the people actually doing that.

They’ve now developed a solution that greatly improves the ability of robots to anticipate the trajectory of humans as they move — allowing robots that typically freeze in the face of anything even vaguely resembling a person walking in their path to continue to operate and move around the flow of human foot traffic.

Researchers managed this by eschewing the usual practice of borrowing from how music and speech processing works for algorithmic prediction, which are much better when it comes to predicting predictable paths of travel, and instead came up with a “partial trajectory” method that references real-time trajectory data with a large library of reference trajectories gathered before.

This is a much better way of anticipating human movement, which is very rarely consistent and involves a lot of stops and starts, even in a factory worker performing the same action repeatedly over thousands of instances.

This could have potential consumer applications too — researchers note that human movement even in the home would be better predicted using this, which could have benefits in terms of robotic long-term in-home care for the elderly, for instance.