Robots with Anticipation and Attention

Artificial intelligence is becoming ubiquitous in our daily life. With increased intelligence, equipment can be made more efficient and more adaptive to changes in its environment. Still, the intelligence of machines is quite weak, and even if there exist vacuum cleaners that vacuums on their own and cars that have intelligent systems that compensate for different weather types this kind of intelligence are hard to compare with human intelligence. One major difference between human intelligence and the intelligent machines of today is perhaps our unique ability to change our behavior as new information is learned, and our ability to generalize behavior to different situations. Most important, we have an ability to use previously learned knowledge to anticipate the outcome of an action even if we have never seen it being performed before. This is an ability that is absolutely essential for an intelligent system. To improve the intelligence of the equipment that is available today, a number of different cognitive functions must be added. One of these is the ability to anticipate actions and choices made by the intelligent system itself, but also those made by agents in its surroundings. The vacuum cleaner should not only learn to clean all corners of the room. It should also learn to anticipate the actions needed. For example when assigned the cleaning the room of the 10 years old boy it may choose to do this every Saturday afternoon, because on Saturday morning the boy is supposed to pick up all his toys laying around the floor. It should learn to anticipate that when it’s raining outside, it may not be a good idea to vacuum in the hallway if someone has just come in from the rain. Instead it if it raining, the best solution may be to start clean the hallway before anyone enters the door completely soaked in water. Another case would be to control the attention of the machine depending on the situation. For example, again the vacuum cleaner should assign more of its attention towards the doorway when working and be prepared to move if anybody enter. The cognitive processes behind anticipation are hard to fully model, but there exist models of partial anticipatory system. We believe that it is necessary to develop anticipatory system to progress further toward truly intelligent systems, and the goal for the next three years is to develop a robot with anticipatory system. As participants in the Mind RACES project, some of Europes top researches in the field will work together to achieve this goal.