How Animated Agents Perform Tasks: Connecting Planning and Manipulation Through Object-Specific Reasoning

Creating animations of a human figure performing a task requires that the agent interact with objects in the environment in a realistic way. Agent-object interaction is not completely specified from a task description alone. In this paper we sketch an architecture for the Object-Specific Reasoner (OSR), an intermediate planning module which tailors high-level plans to the specifics of the agent and objects. As plans are elaborated, the OSR generates a sequence of motion directives which are ultimately executed by a simulator. Descriptions of failures can be used to identify possible tools for the agent to use. An Object-Specific Reasoner is necessary in a system which allows an agent, equipped with a set of action behaviors, to interact in a semiautonomous fashion with the world.