Stay up to date

A new approach for robots: learning by imitating humans

A new approach for robots: learning by imitating humans

Artificial intelligence is already extensively used within the Volkswagen Group, especially when it comes to production. But how can a robot learn from people?

The answer is (apparently) simple: by imitating their movements. And in this case, the tool adopted for the training is certainly innovative. It is a “sensor jacket” equipped with nine sensors positioned on the hands, arms, shoulders and chest. This technological solution was developed by the German start up Wandelbots, and is currently being tested at the Transparent Factory in Dresden. Here, in a dedicated area right next to the production line where the more “traditional” robots work, a technician is showing the machine the movements to be replicated.

Reproducing the movements

And the robot reacts immediately, imitating every single movement. The sensors attached to the jacket register the wearer’s movements in real time and send the data in wireless form to a computer, which processes them at lightning speed to control the robot. Everything takes place in a matter of nanoseconds: when the technician raises his arm, so too does the robot - but the movement is not exactly the same, because the “machine-learning” software programmes the movements in such a way as to optimise the robot’s performance. The more often it is “shown” a movement via the jacket, the better and more reliable it becomes.

A mechanical hand

The initial tasks assigned to the robot are the installation of loudspeakers in the doors or of the power window controls and the application of adhesive pads. For the first task, for example, a human assembly worker actually needs three hands: one to hold the speaker, one to hold the pliers and one to place the rivet on the pliers. Now the robot holds the loudspeaker – and in the future will install it entirely on its own.

Time to learn

Each robot needs a different amount of time to learn the movements, depending on the complexity of the task. It can learn limited and repetitive movements in a few minutes, but it might need several hours for more challenging tasks. The goal is clear though: within a few years, the idea is to have small industrial robots waiting every morning at reception or directly on the production line for the employees to arrive, quickly learn what to do and then giving the human workers time for other responsibilities.

What is already clear is that the robots are not expected to replace humans, but rather to support them efficiently: ideally, every employee who works at an assembly line should be able to teach their “robot colleague” something new every day. – Volkswagen AG

You may like