grounding semantic categories in behavioral interactions: experiments with 100 objects
Author(s): Jivko Sinapov, Connor Schenck, Kerrick Staley, Vladimir Sukhoy, Alexander Stoytchev
Venue: Robotics and Autonomous Systems
Year Published: 2012
Keywords: visual perception, manipulation
Expert Opinion: Interactive perception and multimodal sensing are fundamental aspects of robotics and robot learning. The ability to execute actions to interact with the environment provides robots with a rich source of information, especially when combined with haptic, vision, and audio feedback. Grounding the object representations in the robot's actions provides the robot with representations that are not only well suited for future manipulation tasks, but that the robot can estimate through autonomous experimentation. The paper also touches on actively selecting actions to quickly reduce uncertainty.