Ozgur Reyhanoglu from Biokido Medical Engineering writes about the role that robotics and machinery can play in healthcare.
Robots are increasingly participating in our lives. It can't be denied that they work superbly with repetitive, single tasks such as assembly lines, yet this means reprogramming each single robot for every varying task.
Likewise, they improve the success rate of a high precision surgery such as a brain surgery, but they always need intensive supervision from a specialist. Despite the incredibly fast improvements in robotic technology, we still need human force to perform everyday jobs that need flexibility and high level of expertise.
Did you know according to The WHO’s Global Strategy on Human Resources for Health: Workforce 2030, human resources shortage can mount up to 9.9 million physicians, nurses and midwives globally by 2030? This is the exact reason for our need for "surgeon robots" that can perform a complicated surgery with minimal supervision. We need robots that can collect blood samples, nurse robots, patient care robots, and diagnostic robots, too.
We are not and we will never be able to assign the adequate number of expert human resources for everyone! This is the situation just for healthcare and it might be the same for numerous other industries. But how we will achieve such robotics?
With this perspective, in Biokido we focus our research on building artificial intelligence for such autonomy, inspired by the human brain. We think a robot should be trained just like a baby figures out how to utilise its limbs to control objects. Here, the question is how does a baby achieve this? Learning process of infants is a difficult topic for scientific investigation.
However, there are pieces of information scattered through studies of developmental psychology, animal studies, and cognitive sciences which we can combine to have a general perspective. Briefly, while vision is a fundamental sense, self-perception, and motor control, tactile sensing and object representations in the brain are trained at the same time resulting in an amazing capability of object manipulation. Haven't you ever realised that a baby surprised by the hand in front of the eyes? The baby examines thoroughly the hand which is inside the viewing area. This is a learning process which will lead to being aware that there exists a limb, a "tool" which can be controlled. Then, a baby always tries to reach and grasp an object inside the viewing area.
Maybe the baby intuitively thinks: "There is something over there, and there is another something that I can control! But why I can't control the other thing? Let's see what I can do!" This sequence leads to the amazing result of the human brain that can train itself with little supervision and finally a whole body that can be trained to collaborate, to invent, to build, to do brain surgery!
How do we teach a child to grasp an object that has been never encountered? For example how do we teach a child to drink water from a soft plastic cup, without spilling? Since the child has the fundamental training to control the limbs starting from the infancy, we just show to grasp from the top sturdy part of the cup with the thumb and the forefinger. That's it! Showed once, and our child is able to drink water without trouble.
This is exactly what we work on to achieve for a robot that is aware of the world and its own useful components. We envision that the key components for such an intelligent robot would be visual and kinaesthetic limbs, hand and finger tracking, visual object tracking, tactile sensing, and kinematic control, which are trained simultaneously with a semi-supervised deep neural network architecture. Such a system would learn to recognise objects’ shapes, feel and usability, recognise its own components and learn to control itself, just like a baby learns to see and control itself.
Then it will be ready to be trained for specific purposes and in most cases, just showing the action will be enough. Such a system would have the potential to induce radical changes not only in the medical industry but also in every other industry. Therefore, in Biokido, starting with machine vision utilising artificial intelligence, lead by our skeleton tracking and body pose estimation algorithms, we take this path in order to contribute to humanity better.
