Robot PR2 can make inferences about what it doesn’t know based upon similar objects and its environment.
Although we’re still a long ways off from actual machine intelligence, we are are continuing to make strides towards robotic behaviors that at least emulate it. A key part of intelligence is to make inferences based on given and prior information and to be able to draw conclusions. This is what we normally call common sense. University of Tokyo and Technische Universität München researchers have collaborated on what they call ‘semantic search’ to mimic this aspect of intelligence in robotics. In the researchers’ example when the robot PR2 is told to get a sandwich but there are none where there should be it will use the information it has at its disposal to decide where sandwiches most likely will be. When faced with a kitchen devoid of sandwiches, the robot treks to Subway to purchase one instead. It can develop knowledge about things it doesn’t know based on things it does know.
A video demo shows PR2 in action: