Ruini, F. 1 , Apel, J. K. 2 , Morse, A. F. 1 , Cangelosi, A. 1 , Ellis, R. 3 , Goslin, J. 3 & Fischer, M. 4
1 Plymouth University, Centre for Robotics and Neural Systems, School of Computing and Mathematics - Plymouth, UK
2 University of Dundee, School of Psychology - Dundee, UK
3 Plymouth University, School of Psychology - Plymouth, UK
4 Universitat Potsdam, Department Psychologie - Potsdam, Germany
The work described herein illustrates the development of a cognitive architecture for humanoid robots. Such architecture is intended to provide the iCub robot - used as the reference platform - with the capability of performing short-term memorisation (and subsequent rehearsal) of instruction sequences in a bio-inspired fashion.
This research takes inspiration from psychological experiments that have been carried out on human subjects. Sit in front of a computer screen displaying a 3-by-3 numbered black grid on a white background surrounded by 8 coloured objects, participants were presented with multicomponent instructions like “Move the mug [critical object] to square 5 [goal square], then move the hammer to square 3, then move the pan to square 8”. After attending to all the instruction components and a go signal, they had to replay the sequence in the correct order, moving the various objects to the expected positions using a mouse. Amongst the various findings identified, these experiments have highlighted that during the memorisation phase pairs of objects/squares are repeatedly visually re-visited by some subjects (ocular rehearsal effect). Furthermore, memorisation strategies based upon ocular rehearsal lead to better performances during the execution phase compared to alternative strategies.
The above experimental setup has been reproduced on the iCub computer simulator, with the virtual robot replicating the ocular behaviours exhibited by the human participants. A cognitive architecture relying on a network of interacting SOMs has been designed to keep track of the instructions heard according to a temporal dynamic affected by both auditory and visual information. The fine-tuning of the cognitive architecture parameters has been achieved employing an evolutionary algorithm.
An accurate description of the cognitive architecture is provided within the paper, along with the main results obtained and a comparison with those gathered from the experiments involving human participants.