Intention inference can be an essential step toward efficient humanrobot interaction. For this purpose, we propose the Intention-Driven Dynamics Model (IDDM) to probabilistically model the generative process of movements that are directed by the intention. The IDDM allows to infer the intention from observed movements using Bayes ’ theorem. The IDDM simultaneously finds a latent state representation of noisy and highdimensional observations, and models the intention-driven dynamics in the latent states. As most robotics applications are subject to real-time constraints, we develop an efficient online algorithm that allows for real-time intention inference. Two human-robot interaction scenarios, i.e., target prediction for robot table tennis and action recognition for interactive humanoid robots, are used to evaluate the performance of our inference algorithm. In both intention inference tasks, the proposed algorithm achieves substantial improvements over support vector machines and Gaussian processes. 1

Abstract—Inference of human intention may be an essential step towards understanding human actions and is hence important for realizing efficient human-robot interaction. In this paper, we propose the Intention-Driven Dynamics Model (IDDM), a latent variable model for inferring unknown human intentions. We train the model based on observed human movements/actions. We introduce an efficient approximate inference algorithm to infer the human’s intention from an ongoing movement. We verify the feasibility of the IDDM in two scenarios, i.e., target inference in robot table tennis and action recognition for interactive humanoid robots. In both tasks, the IDDM achieves substantial improvements over state-of-the-art regression and classification. I.

Theabilitytolearnandadaptwhenplayingagainstanadaptive opponent requires the ability to predict the opponent’s behavior. Capturing any changes in the opponent’s behavior during a sequence of plays is critical to achieve positive outcomes in such an environment. We identify two new requirements that we suggest are essentialforagentsthatlearninadaptiveenvironments. These requirements are dictated by the fact that repeated interactions in practice have to be limited and that the opponent can rapidly change strategy through the sequence of interactions. We believe that building intelligent agents that can survive in environments with such requirements will lead to wider deployment of learning agents. We propose a novel algorithm that is able to learn and adapt rapidly to an opponent even when the number of interactions is limited and the opponentis adaptingquickly by changing its strategy. The context we use for the experimental work is two player normal form games. We compare the performance of an agent using our algorithm against agents using existing multiagent learning algorithms. 1.

Abstract not found

Given a prediction of opponent behavior in a general-sum two-player normal form game, it is difficult to se-lect a strategy that balances the opportunity to use the prediction to inform one’s action with the risk of be-coming predictable. We propose Restricted Stackelberg Response with Safety (RSRS), a novel way of generat-ing such a strategy. RSRS uses an r-safe Stackelberg equilibrium in a modified game, which is created to reflect the assumption that the prediction might be in-accurate. With appropriate parameter selection, RSRS produces strategies that can play well against the pre-diction, respond well against a best-responding oppo-nent, or guard against worst-case outcomes. We de-scribe an algorithm that selects appropriate parameter values, which we have tested on multiple general-sum games, comparing its performance to that of other algo-rithms.