Learning regularities with a constructivist agent
AAMAS '06 Proceedings of the fifth international joint conference on Autonomous agents and multiagent systems
Review: learning like a baby: A survey of artificial intelligence approaches
The Knowledge Engineering Review
| Hi-index | 0.00 |
Autonomous robots are used more and more in remote and inaccessible places where they cannot be easily repaired if damaged or improperly programmed. A system is needed that allows these robots to repair themselves by recovering gracefully from damage and adapting to unforeseen changes. Newborn infants employ such a system to adapt to a new and dynamic world by building a hierarchical representation of their environment. This model allows them to respond robustly to changes by falling back to an earlier stage of knowledge, rather than failing completely. A computational model that replicates these phenomena in infants would afford a mobile robot the same adaptability and robustness that infants have. This dissertation presents such a model, the Constructivist Learning Architecture (CLA), that builds a hierarchical knowledge base using a set of interconnected self-organizing learning modules. The dissertation then demonstrates that CLA (1) replicates current studies in infant cognitive development, (2) builds sensorimotor schemas for robot control, (3) learns a goal-directed task from delayed rewards, and (4) can fall back and recover gracefully from damage. CLA is a new approach to robot control that allows robots to recover from damage or adapt to unforeseen changes in the environment. CLA is also a new approach to cognitive modeling that can be used to better understand how people learn for their environment in infancy and adulthood.