In our research on Commonsense reasoning, we have found that an especially important kind of knowledge is knowledge about human goals. Especially when applying Commonsense reasoning to interface agents, we need to recognize goals from user actions (plan recognition), and generate sequences of actions that implement goals (planning). We also often need to answer more general questions about the situations in which goals occur, such as when and where a particular goal might be likely, or how long it is likely to take to achieve. In past work on Commonsense knowledge acquisition, users have been directly asked for such information. Recently, however, another approach has emerged—to entice users into playing games where supplying the knowledge is the means to scoring well in the game, thus motivating the players. This approach has been pioneered by Luis von Ahn and his colleagues, who refer to it as Human Computation. Common Consensus is a fun, self-sustaining web-based game, that both collects and validates Commonsense knowledge about everyday goals. It is based on the structure of the TV game show Family Feud. A small user study showed that users find the game fun, knowledge quality is very good, and the rate of knowledge collection is rapid.

Everyday social interactions are heavily influenced by our snap judgments about others ’ goals. Even young infants can infer the goals of intentional agents from observing how they interact with objects and other agents in their environment: e.g., that one agent is ‘helping ’ or ‘hindering ’ another’s attempt to get up a hill or open a box. We propose a model for how people can infer these social goals from actions, based on inverse planning in multiagent Markov decision problems (MDPs). The model infers the goal most likely to be driving an agent’s behavior by assuming the agent acts approximately rationally given environmental constraints and its model of other agents present. We also present behavioral evidence in support of this model over a simpler, perceptual cue-based alternative. 1

This chapter reviews several approaches to the problem of learning by imitation in robotics. We start by describing several cognitive processes identified in the literature as necessary for imitation. We then proceed by surveying different approaches to this problem, placing particular emphasys on methods whereby an agent first learns about its own body dynamics by means of self-exploration and then uses this knowledge about its own body to recognize the actions being performed by other agents. This general approach is related to the motor theory of perception, particularly to the mirror neurons found in primates. We distinguish three fundamental classes of methods, corresponding to three abstraction levels at which imitation can be addressed. As such, the methods surveyed herein exhibit behaviors that range from raw sensory-motor trajectory matching to high-level abstract task replication. We also discuss the impact that knowledge about the world and/or the demonstrator can have on the particular behaviors exhibited.

Everyday human interaction relies on making inferences about social goals: goals that an intentional agent adopts in relation to another agent, such as “chasing”, “fleeing”, “approaching”, “avoiding”, “helping ” or “hindering”. We present a computational model of social goal inference that takes as input observations of multiple agents moving in some environmental context. The model infers a social goal for each agent that is most likely to have given rise to that agent’s observed actions, under an intuitive theory that expects agents to act approximately rationally. We provide evidence for our theory-based approach over a simpler bottom-up motion cue-based approach in a behavioral experiment designed to distinguish the two accounts.

Abstract — Visibility in architectural layouts affects human navigation, so a suitable representation of visibility context is useful in understanding human activity. Motivated by studies of spatial behavior, we use a set of features from visibility analysis to represent spatial context in the interpretation of human activity. An agent’s goal, belief about the world, trajectory and visible layout are considered to be random variables that evolve with time during the agent’s movement, and are modeled in a Bayesian framework. We design a search-based task in a sprite-world, and compare the results of our framework to those of human subject experiments. Our findings confirm that knowledge of spatial layout improves human interpretations of the trajectories (implying that visibility context is useful in this task). Since our framework demonstrates performance close to that of human subjects with knowledge of spatial layout, our findings confirm that our model makes adequate use of visibility context. In addition, the representation we use for visibility context allows our model to generalize well when presented with new scenes. I.

In this paper we propose a computational model for learning from demonstration. By adequate adjustment of a few parameters, our model is able to produce di erent learning behaviours, taking into account di erent elements of the demonstration. In particular, our model takes into consideration the actions of the demonstrator, its effects on the environment/surroundings, the demonstrator's inferred goals, and the interests and preferences of the learner itself. We present results where we show that our model can reproduce (in simulation) several well-known results from standard experimental paradigms in developmental psychology and also an application to a real robotic imitation learning task.

Abstract not found

Comprehension of goal-directed, intentional motion is an important but understudied visual function. To study it, we created a two-dimensional virtual environment populated by independently-programmed autonomous virtual agents, which navigate the environment, collecting food and competing with one another. Their behavior is modulated by a small number of distinct “mental states”: exploring, gathering food, attacking, and fleeing. In two experiments, we studied subjects ’ ability to detect and classify the agents ’ continually changing mental states on the basis of their motions and interactions. Our analyses compared subjects ’ classifications to the ground truth state occupied by the observed agent’s autonomous program. Although the true mental state is inherently hidden and must be inferred, subjects showed both high validity (correlation with ground truth) and high reliability (correlation with one another). The data provide intriguing evidence about the factors that influence estimates of mental state—a key step towards a true “psychophysics of intention.”

We describe a generative Bayesian model for action understanding in which inverse-forward internal model pairs are considered ‘hypotheses ’ of plausible action goals that are explored in parallel via an approximate inference mechanism based on sequential Monte Carlo methods. The reenactment of internal model pairs can be considered a form of motor simulation, which supports both perceptual prediction and action understanding at the goal level. However, this procedure is generally considered to be computationally inefficient. We present a model that dynamically reallocates computational resources to more accurate internal models depending on both the available prior information and the prediction error of the inverse-forward models, and which leads to successful action recognition. We present experimental results that test the robustness and efficiency of our model in real-world scenarios. 1

views and conclusions contained in this document are those of the authors and should not be interpreted as