We are going without hands
Paralyzed people drove a wheelchair with the power of thought
Nadezhda Chekasina, N+1
Three people with tetraplegia were able to control a wheelchair accurately enough with the power of thought to drive it around a room with furniture and medical equipment. The participants imagined moving their arms or legs, and the chair turned right or left depending on the signal. The control was carried out through the connection of the computer behind the chair with the EEG helmet on the participant's head. The results of the experiment are published in iScience (Tonin et al., Learning to control a BMI-driven wheelchair for people with severe tetraplegia).
Technologies that allow connecting the brain to a computer are actively developing, and especially neural interfaces are important when it comes to people with disabilities — the introduction of neuroprostheses into everyday life helps to ensure the mobility of people with disabilities. Some studies have already shown the success of using neural interfaces for prosthetic arms, exoskeletons, telepresence robots, as well as mind-controlled wheelchairs. When developing such chairs, both invasive and non-invasive methods are used, which are considered safer. With an invasive connection, electrodes are implanted into the patient's brain, and, like any operation, this method carries certain risks and possible complications. With non-invasive methods, scientists use electroencephalography to record brain activity signals. For example, to control a wheelchair, a person focuses on a flickering light in a certain place. Next, the interface reads and decodes brain signals from the electrodes, converts them into a movement command and the chair goes to the place of flickering. But this method also has its drawbacks, for example, eye fatigue or splitting of attention. In addition, although the skull and skin separating the brain cells from the electrodes are conductors, the signal can be distorted.
A group of scientists from the UK, Germany, Italy, the USA and Switzerland led by Jose del R. Millán from the University of Texas at Austin decided to test the hypothesis that it is very important to control the chair in a non-invasive way, how well the patient was able to learn to control the neurointerface. They took three people with spinal cord injury and paralysis of all four limbs and trained them to operate a modified electric wheelchair. At first, they were trained for a considerable time to interact with the neurointerface. Each participant wore an EEG helmet with 31 electrodes that read signals from the area of the brain that regulates movement. These signals were transmitted to a laptop mounted on the back of a wheelchair, where robotic intelligence transformed them into wheel movements. Turning to the left corresponded to the representation of movement with both feet, and turning to the right — with both hands. Otherwise, the wheelchair was moving forward. The screen displayed an improvised steering wheel, an arrow in the center of which pointed in the right direction. If the participant managed to send the signal correctly, the wheels of the chair turned in the right direction. Gradually, the participants got used to controlling the chair without direction hints in the form of arrows.
The training was conducted three times a week: the first participant was engaged for five months, the second — three and the third — two months. In the first lessons, all participants demonstrated low control accuracy — from 43 to 55 percent. The first participant showed himself best in training — at the last session he achieved 95 percent control. The third participant also consistently demonstrated an improvement in management skills (98.3 percent), up to the 7th session, but then his indicators fell (74 percent at the last session). Researchers attribute this to the replacement of the decoder in the participant in recent sessions. The second participant had no obvious success, although he showed stable command of the interface at an average accuracy of 68 percent. The scientists also noticed that during the training, brain signals meaning "left" and "right" became more distinct in participants 1 and 3.
Then the researchers decided to test how the participants would be able to control the chairs in real conditions — a room 15 by 7 meters in the hospital, where there were beds, a screen and medical equipment. Participants had to go around the room and pass four control points, including a U-turn and the need to enter a narrow corridor. The first participant passed 29 experimental sessions, and the second and third participants — 11 each. Of the three participants, only the first and third were able to complete the route and all control points. The first participant successfully passed the last checkpoint in 80 percent of cases, the average time to complete the distance was about 4 minutes. The third participant completed the route in about 6 minutes and was able to reach the fourth checkpoint in 20 percent of cases. But the second participant spent about 5 minutes to reach the third checkpoint and passed it in 60 percent of cases, but he could not pass the last point once and complete the route.
Since the control was carried out with the joint work of the human mind and robotic intelligence, the scientists also assessed how successfully the participants would have completed the route without robot assistance (their predictions of the success of the route with the robot coincided with the real results). It turned out that for the first participant, the overall average success rate of the route would have decreased from 95 percent to 82, for the second — from 45 percent to 22.5, and for the third from 73.3 to 32.5 percent. Researchers pay attention to the high accuracy of wheelchair control for people with limb paralysis not in the laboratory, but in real conditions. They also note that the joint management of the wheelchair makes it possible to achieve efficiency and ease of use, especially if the indications for training interaction with the neurointeface were not high enough (as in the second participant). Despite a number of limitations in the study, scientists believe that the use of mutual learning between a person and a decoder, as well as the introduction of robotic intelligence into management, allows for better results in non-invasive wheelchair management.
You can see how one of the participants goes along a given route here.
Intel has previously shown a wheelchair that can be controlled using facial expressions — the system is able to recognize up to 10 grimaces and allows each to assign its own control team.
Portal "Eternal youth" http://vechnayamolodost.ru