Abstract not found

This review investigates two recent developments in artificial intelligence and neural computation: learning from imitation and the development of humanoid robots. It is postulated that the study of imitation learning offers a promising route to gain new insights into mechanisms of perceptual motor control that could ultimately lead to the creation of autonomous humanoid robots. Imitation learning focuses on three important issues: efficient motor learning, the connection between action and perception, and modular motor control in the form of movement primitives. It is reviewed here how research on representations of, and functional connections between, action and perception have contributed to our understanding of motor acts of other beings. The recent discovery that some areas in the primate brain are active during both movement perception and execution has provided a hypothetical neural basis of imitation. Computational approaches to imitation learning are also described, initially from the perspective of traditional AI and robotics, but also from the perspective of neural network models and statistical-learning research. Parallels and differences between biological and computational approaches to imitation are highlighted and an overview of current projects that actually employ imitation learning for humanoid robots is given.

We use simple learning models to track the behavior observed in experiments concerning three extensive form games with similar perfect equilibria. In only two of the games does observed behavior approach the perfect equilibrium as players gain experience. We examine a family of learning models which possess some of the robust properties of learning noted in the psychology literature. The intermediate term predictions of these models track well the observed behavior in all three games, even though the models considered differ in their very long term predictions. We argue that for predicting observed behavior the intermediate term predictions of dynamic learning models may be even more important than their asymptotic properties.

A briefer version of this paper with the same name has been published in

Abstract not found

Imitation is a powerful mechanism whereby knowledge may be transferred between agents (both biological and artificial). Key problems on the topic of imitation have emerged in various areas close to artificial intelligence, including the cognitive and social sciences, animal behavior, robotics, human--computer interaction, embodied intelligence, software engineering, programming by example and machine learning. Artificial systems used to study imitation can both test models of imitation derived from observational or neurobiological data on imitation in animals and then apply them to different kinds of nonbiological systems ranging from robots to software agents. A crucial problem in imitation is the correspondence problem, mapping action sequences of the demonstrator and the imitator agent. This problem becomes particularly obvious when the two agents do not share the same embodiment and affordances. This paper describes a new general imitation mechanism called Action Learning for Imitation via Correspondence between embodiments (ALICE) that specifically addresses the correspondence problem. The mechanism is implemented and its efficacy illustrated on the "chessworld" testbed that was created to study imitation from an agent-based perspective, i.e., by a particular agent in a particular environment.

We propose Action-Reaction Learning as an approach for analyzing and synthesizing human behaviour. This paradigm uncovers causal mappings between past and future events or between an action and its reaction by observing time sequences. We apply this method to analyze human interaction and to subsequently synthesize human behaviour. Using a time series of perceptual measurements, a system automatically uncovers correlations between past gestures from one human participant (an action) and a subsequent gesture(areaction) from another participant. A probabilistic model is trainedfrom data of the human interaction using a novel estimation technique, Conditional Expectation Maximization (CEM). The estimation uses general bounding and maximization to monotonically find the maximum conditional likelihood solution. The learning system drives a graphical interactive character which probabilistically predicts a likely response to a user's behaviour and performs it interactively. Thus, after analyzing human interaction in a pair of participants, the system is able to replace one of them and interact with a single remaining user. 1

This paper investigates belief learning. Unlike other investigators who have been forced to use observable proxies to approximate unobserved beliefs, we have, using a belief elicitation procedure (proper scoring rule), elicited subject beliefs directly. As a result we were able to perform a more direct test of the proposition that people behave in a manner consistent with belief learning. What we find is interesting. First...

Introduction This chapter presents the agent-based perspective on imitation. In this perspective, imitation is best considered as the behavior of an autonomous agent in relation to its environment, including other autonomous agents. We argue that such a perspective helps unfold the full potential of research on imitation and helps in identifying challenging and important research issues. We first explain the agent-based perspective and then discuss it in the context of particular research issues in studies with animals and artifacts, with reference to chapters presented in this book. At the end of the chapter we briefly introduce the individual contributions to this book and provide a roadmap that helps the reader in navigating through the exciting and highly interwoven themes that are presented in this book. In order to focus discussions, we explain the agent-based perspective with particular consideration of the correspondence

This paper studies decision-making with rules of thumb in the context of dynamic decision problems and compares it to dynamic programming. A rule is a fixed mapping from a subset of states into actions. Rules are compared by averaging over past experiences. This can lead to favoring rules which are only applicable in good states. Correcting this good state bias requires solving the dynamic program. We provide a general framework and characterize the asymptotic properties. We apply it to provide a candidate explanation for the sensitivity of consumption to transitory income.