Perception and Perspective in Robotics
Abstract
To a robot, the world is a sea of ambiguity, in which it will sink or swim depending on the robustness of its perceptual abilities. But robust machine perception has proven difficult to achieve. This paper argues that robots must be given not just particular perceptual competencies, but the tools to forge those competencies out of raw physical experiences. Three important tools for extending a robot's perceptual abilities whose importance has been recognized individually are related and brought together. The first is active perception, in which the robot employs motor action to reliably perceive properties of the world that it otherwise could not. The second is development, in which experience is used to improve perception. The third is interpersonal influences, in which the robot's percepts are guided by those of an external agent. Examples are given for object segmentation, object recognition, and orientation sensitivity; initial work on action understanding also is described. This work is implemented on two robots, "Cog" and "Kismet." Cog is an upper torso humanoid that has previously been applied to tasks such as visually guided pointing and rhythmic operations like turning a crank or driving a slinky. Kismet is an infant-like robot whose form and behavior are designed to elicit nurturing responses from humans. It is essentially an active vision head augmented with expressive facial features so that it can both send and receive human-like social cues.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2003
- Accession Number
- ADA434752
Entities
People
- Paul Fitzpatrick
Organizations
- Massachusetts Institute of Technology