We present a spatial system called Specialized Egocentrically Coordinated Spaces embedded in an embodied cognitive architecture (ACT-R Embodied). We show how the spatial system works by modeling two different developmental findings: gaze-following and Level 1 perspective taking. The gaze-following model is based on an experiment by Corkum and Moore (1998), whereas the Level 1 visual perspective-taking model is based on an experiment by Moll and Tomasello (2006). The models run on an embodied robotic system.

We report the results of an experiment in which human subjects were trained to perform a perceptual matching task. Subjects were asked to manipulate comparison objects until they matched target objects using the fewest manipulations possible. An unusual feature of the experimental task is that efficient performance requires an understanding of the hidden or latent causal structure governing the relationships between actions and perceptual outcomes. We use two benchmarks to evaluate the quality of subjects’ learning. One benchmark is based on (...) optimal performance as calculated by a dynamic programming procedure. The other is based on an adaptive computational agent that uses a reinforcement-learning method known as Q-learning to learn to perform the task. Our analyses suggest that subjects were successful learners. In particular, they learned to perform the perceptual matching task in a near-optimal manner (i.e., using a small number of manipulations) at the end of training. Subjects were able to achieve near-optimal performance because they learned, at least partially, the causal structure underlying the task. In addition, subjects’ performances were broadly consistent with those of model-based reinforcement-learning agents that built and used internal models of how their actions influenced the external environment. We hypothesize that people will achieve near-optimal performances on tasks requiring sequences of action—especially sensorimotor tasks with underlying latent causal structures—when they can detect the effects of their actions on the environment, and when they can represent and reason about these effects using an internal mental model. (shrink)

Reinforcement learning (RL) models of decision-making cannot account for human decisions in the absence of prior reward or punishment. We propose a mechanism for choosing among available options based on goal-option association strengths, where association strengths between objects represent previously experienced object proximity. The proposed mechanism, Goal-Proximity Decision-making (GPD), is implemented within the ACT-R cognitive framework. GPD is found to be more efficient than RL in three maze-navigation simulations. GPD advantages over RL seem to grow as task difficulty is increased. (...) An experiment is presented where participants are asked to make choices in the absence of prior reward. GPD captures human performance in this experiment better than RL. (shrink)

Successfully explaining and replicating the complexity and generality of human and animal learning will require the integration of a variety of learning mechanisms. Here, we introduce a computational model which integrates associative learning (AL) and reinforcement learning (RL). We contrast the integrated model with standalone AL and RL models in three simulation studies. First, a synthetic grid‐navigation task is employed to highlight performance advantages for the integrated model in an environment where the reward structure is both diverse and dynamic. The (...) second and third simulations contrast the performances of the three models in behavioral experiments, demonstrating advantages for the integrated model in accounting for behavioral data. (shrink)

Cognitive architectures are theories of cognition that try to capture the essential representations and mechanisms that underlie cognition. Research in cognitive architectures has gradually moved from a focus on the functional capabilities of architectures to the ability to model the details of human behavior, and, more recently, brain activity. Although there are many different architectures, they share many identical or similar mechanisms, permitting possible future convergence. In judging the quality of a particular cognitive model, it is pertinent to not just (...) judge its fit to the experimental data but also its simplicity and ability to make predictions. (shrink)

Reinforcement learning approaches to cognitive modeling represent task acquisition as learning to choose the sequence of steps that accomplishes the task while maximizing a reward. However, an apparently unrecognized problem for modelers is choosing when, what, and how much to reward; that is, when (the moment: end of trial, subtask, or some other interval of task performance), what (the objective function: e.g., performance time or performance accuracy), and how much (the magnitude: with binary, categorical, or continuous values). In this article, (...) we explore the problem space of these three parameters in the context of a task whose completion entails some combination of 36 state–action pairs, where all intermediate states (i.e., after the initial state and prior to the end state) represent progressive but partial completion of the task. Different choices produce profoundly different learning paths and outcomes, with the strongest effect for moment. Unfortunately, there is little discussion in the literature of the effect of such choices. This absence is disappointing, as the choice of when, what, and how much needs to be made by a modeler for every learning model. (shrink)

Human control of action in routine situations involves a flexible interplay between (a) task-dependent serial ordering constraints; (b) top-down, or intentional, control processes; and (c) bottom-up, or environmentally triggered, affordances. In addition, the interaction between these influences is modulated by learning mechanisms that, over time, appear to reduce the need for top-down control processes while still allowing those processes to intervene at any point if necessary or if desired. We present a model of the acquisition and control of goal-directed action (...) that goes beyond existing models by operationalizing an interface between two putative systems—a routine and a non-routine system—thereby demonstrating how explicitly represented goals can interact with the emergent task representations that develop through learning in the routine system. The gradual emergence of task representations offers an explanation for the transfer of control with experience from the non-routine goal-based system to the routine system. At the same time it allows action selection to be sensitive both to environmental triggers and to biasing from multiple levels within the goal system. (shrink)

We investigate how Reinforcement Learning methods can be used to solve the production selection and production ordering problem in ACT‐R. We focus on four algorithms from the Q learning family, tabular Q and three versions of Deep Q Networks, as well as the ACT‐R utility learning algorithm, which provides a baseline for the Q algorithms. We compare the performance of these five algorithms in a range of lexical decision tasks framed as sequential decision problems.