The verbal instructions a person gives to help someone find a desired object are very difficult for a robot to use (the cup over near the couch or the brush next to the red toothbrush). These types of commands require the robot to understand everyday human language and the objects it describes at a level well beyond the state of the art in language recognition and object perception.
"We humans naturally point at things but we aren't very accurate, so we use the context of the situation or verbal cues to clarify which object is important," said Kemp, an assistant professor in the Wallace H. Coulter Department of Biomedical Engineering at Georgia Tech and Emory. "Robots have some ability to retrieve specific, predefined objects, such as a soda can, but retrieving generic everyday objects has been a challenge for robots."
The laser pointer interface and methods developed by Kemp's team overcome this challenge by providing a direct way for people to communicate the location of interest to El-E and complimentary methods that enable El-E to pick up an object found at this location. Through these innovations, El-E can retrieve objects without understanding what the object is or what it's called.
In addition to the laser pointer interface, El-E uses another approach to simplify its task. Indoors, objects are usually found on smooth, flat surfaces with uniform appearance, such as floors, tables, and shelves. Kemp's team designed El-E to take advantage of this common structure.
Regardless of the height, El-E uses the same strategies to localize and pick up the object by elevating its arm and sensors to match the height of the object's location. The robot's ability to reach objects both from the floor and shelves is particularly important for patients with mobility impairments since these locations can be difficult to reach, Kemp said.
El-E uses a custom-built camera that is omni-directional to see most
|Contact: Megan McRainey|
Georgia Institute of Technology