Robot Uses “Chaos Filter” To Help Recognize Changing Places

Writing by Evan Ackerman on Wednesday, 25 of February , 2009 at 5:50 am

Autonomous robot navigation in complex environments is tricky. When I say “complex,” what I really mean is complex environments that may change. For example, a software algorithm can compare two very complex scenes and determine if they are more or less identical, even from different angles or under different lighting conditions. But, say your complex scene is a busy city street. The exact same perspective at the exact same time each day might be radically different, as there are different cars in the street, different people on the sidewalk, and different signs in store windows. Even though the content of the scene is different, the scene itself is the same, and this is a hard thing for robots that rely on sensor similarity analysis to understand.

Researchers from the Oxford Mobile Robots Group have developed a mapping system called FabMap (Fast Appearance Based Mapping) that compares two scenes the same way that people do to determine whether or not they’re looking at the same place. Instead of analyzing every detail of a scene, humans group things into units, and figure out how different scenes are based on that. For example, when comparing an occupied and vacant driveway scene, a robot might say, “this scene is missing a red boxy object on top of some round objects that took up a significant percentage of the previous scene, therefore this scene must be different.” A human would know that despite all of the different details, there is only actually one difference in the scene: there’s no car in it. This is what FabMap does: it allows robots to group objects into units and recognize when a single unit has changed as opposed to many discrete features. This attention to specific units rather than discrete features also makes the system much better at recognizing when two scenes which look almost the same, aren’t.

In the image pairs above, green circles represent similarities while red circles represent changes. Even when there are significant differences like the absence or presence of a person, the software is still able to recognize major similar units in the images and match them with a high probability.

The obvious advantage to this technique is that it doesn’t require any complex hardware or external guidance (like a GPS signal), meaning that it works just as well indoors. Plus, it’s more adaptable than GPS outdoors, providing actual information about the environment rather than just a set of coordinates.

[ FabMap ] VIA [ New Scientist ]

Comments (4)

Category: Artificial Intelligence

• Add this post to
• Del.icio.us -
• Digg

Name (required)

Mail (will not be published) (required)

Website

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>