A research group at DTU Electrical Engineering headed by Assistant Professor Silvia Tolu has developed a new bio-inspired control system for robots based on the human cerebellum. The cerebellum controls our movements and adjusts them to the surroundings, both in relation to activating the right muscles when—for example—we are to walk and in relation to making the necessary adjustments to enable us to continue walking even though the surface changes or rocks like the deck of a ship.
“We’ve translated the many processes of the cerebellum into an ‘engineering system’ in which we’ve combined the biological mechanisms with conventional control techniques to control robots. In this way, we can enable the robot to master movements while also adapting to changes in the surroundings,” says Silvia Tolu.
One of the tests conducted by the researchers involved a musculoskeletal mouse robot. It was made to walk on two rolling belts, one for the two legs on the left side of the robot and one for the two legs on the right side. The speed on the right belt was then increased, and the robot automatically adjusted the walking parameters to stabilize the movement. These provisional findings confirm the results achieved under similar tests with real mice and humans.
Silvia Tolu has no doubt that robots with more soft and adjusted movements will be a benefit for future use of robots. It can—for example—increase safety when robots and humans work together side by side, so that not only humans are to adjust to the movements of the robot—but also vice versa—if needed.
Robots learn new things themselves
Basing the operating system on an imitation of the ability of the cerebellum to remember movements opens up completely new possibilities for the use of robots.
“Using machine learning techniques, we’ve succeeded in enabling robots to receive a message and then find out for themselves how to solve the task. Here, we utilize the ability of the cerebellum to learn from its mistakes so that it achieves the desired action,” says Silvia Tolu.
“Think—for example—of a small child learning how to walk. Here, a lot of practice is needed in order to succeed—and along the way the child’s cerebellum learns what brings the child closer to the goal and what prevents it. We’ve added this ability to the robot, so that its motor becomes self-learning,” says Silvia Tolu.
One of the researchers’ tests has shown that a self-learning robot arm makes several errors to begin with when it is asked to move in an 8-figure form. However, the total number of errors made by the robot controlled by a simple cerebellar functionality quickly decreased significantly. After a short robot learning period, the researchers thus measured 30% fewer errors in the 8-figures than in the robots controlled by a conventional operating system.
Rehabilitation of patients with Parkinson’s disease
The next step for Silvia Tolus’s research is to examine how robots can be used in the understanding and treatment of brain diseases such as Parkinson’s disease.
“This may—for example—be in connection with rehabilitation, where robots can contribute to replacing or strengthening the functions of the cerebellum in the patients. To meet this objective, the operating system must be extended to comprise other areas of the brain, especially the basal ganglia. The interaction between the cerebellum and the basal ganglia is essential in understanding Parkinson’s disease,” explains Silvia Tolu.
Knowledge of the brain has been important in enabling the DTU researchers—as some of the first in the world—to make a new bio-inspired operating system for robots based on the functions of the cerebellum. However, Silvia Tolu’s tests with robots also provide useful knowledge to brain researchers which can help confirm or refute theses about the functions of the brain. If the robot is unable to act as expected under given circumstances, this can contribute to new hypotheses or insight into the human brain.