Integrating Language and Cognition in Grounded Adaptive Agents

EOARD Grant # 053060

 

 

 

Results:

The following is a summary of the experimental and modelling results obtained so far:

 

             MFT - (Modelling Field Theory):

(All the following results were implemented using Matlab but were then later transferred, for future use, in C++ .NET )

  1. Replication of the MFT algorithm presented at KIMAS'05. view results

  1. Scale up of the MFT algorithm (Stage 1 - number of Concept Models equals number of Objects) view results

  1. Scale up of the MFT algorithm (Stage 2 - number of Concept-Models is not equal to the number of Objects)      view results

  1. Scale up of the MFT algorithm (Stage 3 - One value for each Concept-Model, but using three values for each object) view results

  1. Scale up of the MFT algorithm (Stage 4 - Using three values for each Concept-Model and three values for each object) view results

  1. Scale up of the MFT algorithm (Stage 5 - One value for each Concept-Model, but a Matrix for each object) view results

  1. Scale up of the MFT algorithm (Stage 6 - One Matrix for each Concept-Model, and One Matrix value for each object) view results

  1. Scale up of the MFT algorithm (Stage 7 - As for the previous stage but using coordinate data from the Robotic Model as an Object) view results

                                        -------------------------------------------------------------

  1. Extension of the Modeling Field Theory neural network for the classification of objects (as seen in ICANN 06)

  • the system is able to dynamically adapt when an additional feature is introduced during learning view results

  • the system can be applied to the classification of action patterns in the context of cognitive robotics view results

  • the system is able to classify multi-feature objects from complex stimulus set view results

  1. MFT_ODE Movie as presented at FUSION 2006 EOARD view results

  1. Scaling up of lexicon and action repertoire in Cognitive Agents: Simulation Details and Results                                                                                                                                                              (KIMAS 07) MFT for the acquisition of language in cognitive robotics.. By using concept-models with multiple sensorimotor modalities, a MFT system can integrate language-specific signals with other internal cognitive representations. view description

  • agents are able to acquire a complex set of actions by building sensorimotor concept-models - view results

  • agents are able to learn a lexicon to describe these objects/actions through a process of cultural learning - view results

  • agents learn actions as basic gestures in order to generate composite actions. view results

     

    •

                                       -------------------------------------------------------------

Available software

 (Click or right click and save as to download .zip file)

 

MFT_ODE