From racing bike to mountain bike

If we know how to ice-skate, we will more rapidly learn how to rollerblade. If we learned to ride a bike on a Dutch bike or on a racing bike, we will rapidly learn how to ride a mountain bike – though the feedback from our muscles and the exact movement sequences are quite different. How do we generalize motion sequences, how do we transfer the ability to ride a bike from one bike to the other? This has now been investigated by scientists around Carsten Mehring from the Bernstein Center for Computational Neuroscience and the University of Freiburg. They managed to show that we learn in a “structured” manner – i.e.: we learn which movement aspects are connected in which way. The work was published in the scientific journal “Current Biology”.

In complex motor activities like riding a bike, we must control many parameters – e.g. the position of arms and legs and the tension of the trunk musculature. If we learn a similar activity, these parameters are also similar. But is this enough to explain how we generalize motion sequences? According to the current study by Mehring and his colleagues, there is more behind it. We learn which control parameters characterize a certain motion class and how they are connected. For example, when we push down the bike’s pedal with the left foot, we move our right foot upwards. And that’s not all: “There are more than 600 muscles in the human body that each need to be more or less contracted  in a coordinated manner, and there are many sensory, visual and haptic responses, which are connected in a certain way for one class of motor tasks, e.g. riding a bike,” says Mehring. “If we switch from one bike to another, we only change the motor aspects that are important to this motion class and we only test parameter combinations that make sense for this motion class.”

In various experiments, Mehring and his colleagues have verified the hypothesis that such a motion principle can be learned. In one experiment, for example, the subjects were asked to move the cursor to a certain point on a computer screen. In this experiment, there was a rotation between the movement of the hand and the movement of the cursor on the computer screen – if the subjects moved the mouse to the right, the cursor moved e.g. diagonally down, and the subjects had to adjust their movement accordingly. From trial to trial the rotation angle changed, thus making it impossible to learn a specific angle of rotation.

Nevertheless, after a couple of trials, the subjects had learned something. If they were now asked to perform the same rotation task several times in a row, they managed to make straight and quick movements much more rapidly than subjects without previous experience. “They had learned the principle that the resulting cursor movement is rotated relative to their own movement – as opposed to being scaled or mirrored, for example” says Mehring. Therefore, they only had to vary a few parameters in order to achieve the right motion sequence. That’s what scientists call “structural learning”. The generalization of motion sequences – as shown by the scientists around Mehring – is based on structural learning.