Teaching robots to collaborate with humans

Robots work faster and longer than humans. They also work in environments where humans would not be safe to perform tasks for example in heavy industries such as car making, but the trade-off is that they can’t be assigned to a completely new task. On the contrary, humans are flexible by nature. Rather than trying to recreate humans, researchers are willing to design robots that are able to work with humans and that can learn from them by mimicking. The European research and innovation project called Collaborate aims to develop innovative and safe human-robot collaborations.

 “To equip robots with such new collaboration skills is a challenging project”, explains Professor Zoe Doulgeri, coordinator of the project. “This way robots will become valuable human assistants, able to understand humans’ intentions, learn from them and work with humans using their abilities of precision and heavy loads manipulation, while leaving the decision-making and movement initiative to humans.” This innovative approach will allow to use robots more effectively, especially for tasks requiring more flexibility. This evolution is crucial to respond to industry needs, as production cycles become shorter and the need to quickly develop new products grows.

Idiap’s touch

“To interact in an effective way with humans, a robot must be able to deal with our very own human nature”, explains Sylvain Calinon, head of the Robots learning & interaction group at Idiap. On the contrary of a robot usually programmed to follow the exact same trajectory, humans will slightly deviate, accelerate or slow down each time they perform a same task. Robots must be able to adapt to such situations. “We work on robots able to learn by imitation”, adds Sylvain. “Thanks to machine learning algorithms, robots can identify a pattern in what we show them and then they can replicate it and adapt themselves if there is a variation.” This feature allows robots to interact with humans in a safer way by avoiding potential collision and gives them much more flexibility to learn new tasks from humans.

In order to transfer research to industry and respond to day to day uses, the project gathers 14 partners across 10 European countries from research, but also industry and services. Collaborate is a three-year project.

Project website: https://collaborate-project.eu