When a robot is tasked to operate in a new environment, it should have the ability to leverage external knowledge sources to acquire common knowledge about its general environment instead of learning everything from scratch. For example, a maintenance robot should have the ability to leverage common knowledge about tools, just as a home robot should have access to knowledge about household items.
Currently, we have contributed a domain-independent framework for generating a context-specific knowledge network for common sense reasoning. Instead of creating general-purpose knowledge base, we studied the extent to which local observations can be leveraged to retrieve relevant semantic information at a scale that is more efficient, and often more computationally tractable. We embedded the reasoning framework within a robot architecture and demonstrated its use in enabling a mobile robot to perform a series of real-world tasks.