|Action Execution, Its Estimation and Learning for a Systems Level Cognitive Architecture|
An agent or robot achieves its goals by interacting with its environment, cyclically choosing and executing suitable actions. Cognitive architectures are considered the control structures of the agent, helping it decide what to do next, while the designs resemble how minds work, be they human, animal, or artificial.
An action execution process is a critical part of an entire cognitive architecture, because the process of generating executable motor commands is not only driven by low-level environmental information, but is also initiated and affected by the agents high-level mental processes. I give a review of the cognitive models of the action execution process as implemented in a set of popular cognitive architectures, and conclude with some general observations regarding the nature of action execution.
Next, I present a cognitive modelthe Sensory Motor System (SMS)for an action execution process, as a new module of the LIDA (for Learning Intelligent Distribution Agent) systems-level cognitive model. A sensorimotor system derived from the subsumption architecture has been implemented into the SMS; and several cognitive neuroscience hypotheses have been incorporated as well.
Inspired by the hypothesis that humans estimate their movements based on their knowledge of the dynamics of the environment, and on actual sensory data (Wolpert, Ghahramani, & Jordan, 1995), I create a model of the estimation process of action execution using SMS in LIDA. Also, based on a recent study in neuroscience (Herzfeld, Vaswani, Marko, & Shadmehr, 2014), I introduce a new factormemory of errorsinto this model of estimation. The historical errors help humans determine the stability of the environment, so as to decide the degree to which knowledge of the environment may affect the estimation.
Learning is significant for for allowing an agent to act more intelligently. I present a new model of sensorimotor learning in LIDA, one that helps an agent properly interact with its environment using past experiences. Following Global Workspace Theory, the primary basis of LIDA, this learning is cued by the agents conscious content, the most salient portion of the agents understanding of the current situation. Furthermore, I add a dynamic learning rate to control the extent to which newly arriving conscious content may affect the learning.
Finally, I introduce an extension of the SMS. This extension allows, and explains, the use of the sensory data, the prime, perceived before a participant starts his or her movement, by the SMS during action execution. Furthermore, this extension allows the replication by a LIDA-based agent, of some human experiments (T. Schmidt, 2002) studying the priming process in motor control.