Can Robots think like humans?

Most children learn geometry very hard, but they are able to catch a ball without calculating its pallet line. So why robots need to think differently? A group of European scientists has developed an artificial cognitive system that learns from experience and observation rather than relying on existing laws and patterns.

Under the direction of Linköping University, Sweden, scientists from the COSPAL project have used a progressive method to make robots identify, identify and interact with objects, especially in love. Random and unexpected situations.

Traditional robot technology relies on the ability of robots to perform complex calculations, such as determining the geometry of an object and predicting its path if it moves. But COSPAL turned this process around, making robots perform tasks based on their own experience and human observation . This test-and-error method can lead to the creation of many automated robots and even enhance our understanding of the human brain.

Michael Felsberg, coordinator of the European-funded COSPAL project, said, 'Gösta Granlund, head of the Computer Vision Laboratory at Linköping University, came up with the idea of ​​action ahead of awareness in learning. This may sound counterintuitive, but it is exactly how people learn. '

He stressed that children 'always test and try everything'. By doing random actions - pointing at this object or touching another object - children will understand the cause-effect and be able to apply this knowledge in the future. By experimenting, children quickly discovered that a ball would roll and could not grasp a hole. Children can learn from observing adults and mimicking their actions, increasing their understanding of the world around them.

Learning like people and learning from people

Robots are recognizing signals and symbols as colors. (Photo: COSPAL)

Applied in the context of the artificial cognitive system (ACS) , this method helps create robots that can learn as much as humans and can learn from people themselves, allowing them to continue performing tasks the case when their environment changes or when an object they are not programmed to recognize is placed in front of them.

'Most artificial intelligence-based ACS systems are quite successful in recognizing objects based on geometry calculations or sources. Some people argue that people also do such calculations to determine something but I don't believe it. I think people are very good at recognizing the geometry of objects from experience. '

The COSPAL group's ACS system seems to want to prove the theory. A robot without pre-programmed geometric knowledge can recognize objects simply from experience, even when the surroundings and the location of the camera it uses to capture image information are altered.

Put the right hook into the right hole

This type of discrimination game used to teach children is used to test this system. Through trial and error and observation, the robot can place cubes into square holes and round hooks into circular holes with an accuracy of 2mm and 2 degrees.'This proves that it can solve geometric problems without having to understand geometry. In fact, I watched my 11-month-old son solve the game: the learning process on babies and robots is very similar. '

A test of the robot's ability to learn from observation involves using a robotic arm to mimic the movement of a human arm. With only 20 to 60 observations, the robotic arm can monitor the movement of the human arm through a false space, avoiding obstacles on the road. In later tests with the same robot, the learning time was greatly reduced, indicating that ACS actually draws memories based on past observations.

In addition, by applying these seemingly logical ideas, the team devised a way to help the robot identify signals and symbols like colors. Instead of using three numbers to represent red, green, and blue as in digital image processing applications, the team captured this system to learn colors from pairs of images and color-coded names. Corresponding references such as red, crimson and dark green in the form of channel coding. Similar to the way that the human brain recognizes colors with selective neuron groups to distinguish green from black, channel coding is a biological-based way of conveying information.

'As humans, we can use the reason to interpret what an object is by the elimination process. For example, we know that if an object has such characteristics, it is something like this and not another. Although this type of machine explanation has been used before, we have developed a progressive version of the object identification process that uses symbolic information and images up to influence. great.' Felsberg said.