Cerno-CAMAL (A Cognitive Robot with vision and touch)


[Picture] Pioneer P3DX in a tight maze world with accompanying Pioneer P2

This new research project makes use of a Pioneer P3DX robot bought using a University of Hull Alumni Development Fund.

The primary aim of this research project is to investigate cognitive architectures and, in particular, their application to Cognitive Robotics. More specifically, the project will investigate how cognitive architectures can reason probabilistically about domain models through perception and learning, and relate their learning to perceptual and domain reasoning. This problem is to be tackled through taking an existing AmigoBot robot from an earlier research project (Robo-CAMAL) with a simplified cognitive architecture (only two layers of reactive and deliberative) and then investigating the shortcomings and drawbacks of the whole implementation. An extended version of the complete cognitive architecture (CAMAL 2008 by D.N.Davis), together with the learning outcomes from the Robo-CAMAL research project, will be used to develop Cerno-CAMAL on the new Pioneer P3DX robot. Thus, this research project will include an implementation (Cerno-CAMAL) which will address some of the theoretical questions regarding perceptual and domain learning in cognitive architectures with the following aims and objectives.

Objectives

  • An investigation into cognitive architectures and their application to Cognitive Robotics
  • Insights into the nature of perceptual and domain learning mechanisms
  • An analysis of perceptual learning using the domain model properties of objects as a guide to their actions
  • An investigation of two common approaches to perceptual learning: Connectionist (Neural Nets) and Probabilistic (Bayesian Belief Nets)
  • The application of CAMAL 2008 to a cognitive robot using the lessons learned from earlier projects
  • A synthesis of a belief reasoning system to implement belief updates within the deliberative layer of Cerno-CAMAL
  • Relating Bayesian metrics to existing Affect metrics used throughout CAMAL 2008

    Aims & Goals

  • Improve our understanding of how perceptual and domain learning occurs in cognitive architectures
  • Develop a perceptual and domain learning system for a cognitive robot
  • Investigate how CAMAL 2008 underpins autonomy, motivation, affect, and learning
  • Develop an adaptive belief revision system to deepen the perceptual and domain reasoning of a cognitive robot
  • Implement a Bayesian belief reasoning system that is compatible with the existing Affect and BDI models used throughout CAMAL 2008
  • Analyze how CAMAL 2008 is able to recognize known objects and relate the behavior of them to existing domain and task models, and extend this ability where domain model assumptions are broken
    Cerno is Latin for perceiving and distinguishing!
    File maintained by Dr D.N.Davis @hull.ac.uk
    Last Updated April 2009