Studying the Coupled Learning of Procedural and Declarative Knowledge in Cognitive Robotics

摘要

Procedural and Declarative knowledge play a key role in cognitive architectures for robots. These types of architectures use the human brain as inspiration to design control structures that allow robots to be fully autonomous, in the sense that their development depends only on their own experience in the environment. The two main components that make up cognitive architectures are models (prediction) and action-selection structures (decision). Models represent the declarative knowledge the robot acquires during its lifetime. On the other hand, action-selection structures represent the procedural knowledge, and its autonomous acquisition depends on the quality of the models that are being learned concurrently. The coupled learning of models and action-selection structures is a key aspect in robot development, and it has been rarely studied in the field. This work aims to start filling this gap by analyzing how these concurrent learning processes affect each other using an evolutionary-based cognitive architecture, the Multilevel Darwinist Brain, in a simulated robotic experiment.