AAAI Publications, The Twenty-Sixth International FLAIRS Conference

Font Size: 
A Framework for the Qualitative Comparison of Diverse Developmental Agents
Lisa B. Soros, Avelino J. Gonzalez

Last modified: 2013-05-19


Developmental agents learn task-nonspecific skills through environmental interactions. The humanlike flexibility of such agents isn’t captured by domain- specific performance metrics. We present a novel framework that complements traditional metrics by allowing cross-domain comparison. The framework considers four properties of developmental agents: the degree of human involvement in design, the length of the agent’s developmental period, the architectural support for acquiring new behaviors, and the tolerated dimensionality of input. This framework is applied to real-world systems in three case studies. We find that our framework allows cross-domain comparison that would not be contributed by traditional quantitative metrics.


Evaluation; Autonomous Mental Development; Developmental Robotics; Epigenetic Robotics

Full Text: PDF