Artificial Intelligence that Learns From Experience

cognitive robot

In the not so distant future, robots may be able to recognize and learn from their experiences. At least that is what European researchers in Sweden are trying to develop. In no time, robots may be learning just like humans do.

A team of European researchers led by a team from Linkoping University in Sweden has developed an artificial cognitive system that learns from experience, by trial and error and observation rather than learning based on predefined rules. The researchers in the COSPAL project made use of an innovative approach to help robots recognize identify as well as interact with objects especially in unforeseen situations.

In the traditional sense, robots do actions by carrying out complex calculations such as measuring the geometry of an object and then measuring the trajectory of an object as it moves. This is unlike human thinking which researchers say are based more on experience and observation rather than complex calculations. A notable example would be how children learn when growing up.

Children, the researchers noted, tests and try everything by performing random actions. They may poke an object, touch it, or throw it around. Through these actions, they come to observe, understand the cause and effect and try to apply that observation in the future. This context has been applied on the Artificial Cognitive system (ACS) that the researchers developed.

This type of approach to artificial intelligence can help create robots that can learn as much as humans do as well as learn from the humans also. This will also allow the robots to perform tasks even through environment changes or when objects that they are not pre-programmed to recognize are placed in front of them.

The COSPAL team’s version of artificial cognitive system (ACS) was able to permit a robot with no pre-programmed geometric knowledge to recognize objects simply from experience. The robot was also able to recognize objects even when its surroundings and the camera position from which it obtained visual information was changed.

The same robot was also made to undergo a shape sorting puzzle that is also used to teach small children about shapes. By trial and error and by means of observation, the robot was also able to place cubes in the square holes and the round pegs in the round holes. The test showed that the robot was also able to solve geometric problems without previous knowledge of geometry.

Source: "Robotic Minds Think Alike?." ScienceDaily 31 March 2008. 28 April 2008.

Leave a Reply

Back to top