Learning task representations from experienced demonstrations

AAAI Conferences

The observations also provide information about objects in the environment relevant to particular skills from the robot's behavior set. Through the same anchoring process these objects are linked to the behaviors that operate on them. The objects do not have any symbols assigned, but they are linked to behaviors through their properties: observed physical features, such as color, size, etc. To summarize, the anchoring process in our teaching by demonstration domain is characterized by: grounding high-level task representations to observations in the form of behavior networks; linking objects to robot behaviors through the objects' observed physical features. Next we describe our behavior architecture and the network representation of robot tasks.


The Character of Human Behavior Representation and its Impact on the Validation Issue

AAAI Conferences

Human behavior representations (HBRs, for short) are a challange for system developers and validation technologies. Authors agree that classical technologies of knowledge processing can not ada equately be applied to HBRs and publish first ideas towards appropriate architectures. In fact, their suggestions are usually driven by a particular application field. To illustrate some characteristics and their impact on validation technologies we introduce a tiny example and outline characteristic features. The upcoming exemplary insights nicly confirm the problems reported by other authors and raise at least two major insights: (1) First research steps should aim at a classification of HBR systems that need different architechtures as well as validation technonlogies.


Behavior Bounding: An Efficient Method for High-Level Behavior Comparison

AAAI Conferences

In this paper, we explore methods for comparing agent behavior with human behavior to assist with validation. Our exploration begins by considering a simple method of behavior comparison. Motivated by shortcomings in this initial approach, we introduce behavior bounding, an automated model-based approach for comparing behavior that is inspired, in part, by Mitchell's Version Spaces. We show that behavior bounding can be used to compactly represent both human and agent behavior. We argue that relatively low amounts of human effort are required to build, maintain, and use the data structures that underlie behavior bounding, and we provide a theoretical basis for these arguments using notions of PAC Learnability. Next, we show empirical results indicating that this approach is effective at identifying differences in certain types of behaviors and that it performs well when compared against our initial benchmark methods. Finally, we demonstrate that behavior bounding can produce information that allows developers to identify and fix problems in an agent's behavior much more efficiently than standard debugging techniques.


Behavior Bounding: An Efficient Method for High-Level Behavior Comparison

Journal of Artificial Intelligence Research

In this paper, we explore methods for comparing agent behavior with human behavior to assist with validation. Our exploration begins by considering a simple method of behavior comparison. Motivated by shortcomings in this initial approach, we introduce behavior bounding, an automated model-based approach for comparing behavior that is inspired, in part, by Mitchell’s Version Spaces. We show that behavior bounding can be used to compactly represent both human and agent behavior. We argue that relatively low amounts of human effort are required to build, maintain, and use the data structures that underlie behavior bounding, and we provide a theoretical basis for these arguments using notions of PAC Learnability. Next, we show empirical results indicating that this approach is effective at identifying differences in certain types of behaviors and that it performs well when compared against our initial benchmark methods. Finally, we demonstrate that behavior bounding can produce information that allows developers to identify and fix problems in an agent’s behavior much more efficiently than standard debugging techniques.


Motor Learning by Imitation

AAAI Conferences

Learning through various forms of imitation is ubiquitous in nature (McFarland 1985, McFarland 1987). Animals imprint, mimic, and imitate adults of their own kind instinctively, often without obtaining direct rewards or successfully achieving the goal of the behavior (McFarland 1985, Gould 1982). The propensity for imitation appears to be innate since imitation is a critical form of learning during development and throughout life. Imitation is defined as the ability to observe and repeat the behavior of another animal or agent, and is one of the principal modes of acquiring new patterns of behavior. "l rue imitation is distinct from mimicry and social facilitation.