Exploring skill generalization with an extra robotic arm for motor augmentation

According to a recent study published in Advanced Intelligent Systems, the brain can adapt to an artificial third arm and use it for simple tasks. This keeps alive the dream of precision mechanics and surgeons for people to deftly use a third arm sometime in the future.

About 20 study participants are learning to use an artificial arm in a laboratory setting. The rudimentary limb, equipped with a clamp at its end, is fixed to a table next to the participants who, while seated, control it using a belt placed on the diaphragm. An exhalation moves the arm forward; an inhalation moves it backward.

The participants practice performing a series of tasks such as grasping blocks, pressing buttons or moving cursors. For the scientists running this project, the goal is to determine to what extent the brain can learn to use a robotic limb in the same way as a natural arm.

In previous studies, the EPFL team had already demonstrated that participants could control virtual arms and point at objects with a simple robotic arm. Now it is one step further as it examines the ability to grasp.

The team led by postdoc David Leal measured a skill that is as banal as it is complex: generalizing tasks. “With a natural limb, we do this automatically,” explains Silvestro Micera, lead author of the study. “If a child learns to grasp a certain object, he or she doesn’t have to learn to do so afterwards for other objects. The brain internalizes the principle of the movement sequence and generalizes it for other objects.”

Multitasking near impossible

According to Micera, if the brain can generalize tasks with an artificial arm, this indicates that it is able to incorporate it—that is, to use it effectively as an integral part of the body. “It’s a clue that suggests that the brain can really control a robotic limb,” he says.

The study shows that generalization does indeed take place. The participants initially practiced moving blocks as quickly as possible using their natural and artificial arms simultaneously. In a second phase, when compared with participants lacking practice, they were able to manipulate various other objects more quickly and more precisely with their natural arms as well as their robotic arm.

In other words, an efficient protocol to induce generalization with a natural limb produced the same effect with the robotic limb.

That said, there is less generalization when the operations requested in the test phase are too far removed from the training phase. This is particularly true in a multitasking context. For example, participants find it difficult to generalize grasping objects with the artificial arm if they have to type on a keyboard at the same time using their hands.

According to Micera, this result suggests that generalization with the artificial limb could be more difficult to achieve and may be limited to performing very similar tasks. And perhaps the training was not optimal either, he adds.

Higher-precision research too invasive

For the time being, relatively few scientists are working on the augmentation of humans with robotic limbs. In the United States and Europe, only a handful of teams are studying the subject, including the integration of artificial fingers. Nevertheless, the promise inherent in this approach remains intriguing.

“You can imagine a lot of professions where additional limbs could be useful. For example, first responders, precision mechanics or surgeons, who would no longer need assistants to pass them their instruments,” says Micera. But he is quick to point out that such applications are still far from being a reality.

The main obstacle lies in the reduced control. Even if improved, the diaphragm control of an artificial arm will remain rudimentary, far from the precision of a natural limb. To clear this hurdle, an invasive interface, such as electrodes in the cortex, could be the long-term solution to translating brain signals into executable commands for the arm.

But this is not possible now. Hence, Micera and his team are limiting themselves to non-invasive devices—currently controlled by the breath and, in the near future, by electrodes placed on the scalp.

For Micera, however, the appeal of such work lies not so much in futuristic scenarios with augmented humans as in a better understanding of the brain and its way of interfacing and building new connections with the body.

“For me, it’s above all a neuroscientific question,” he explains. “By better understanding how to improve and speed up training with an artificial arm, we may gain useful insights for rehabilitation, for example with patients who are paralyzed after a stroke.”