User preference-based integrated multi-memory neural model for improving the cognitive abilities of autonomous robots

Abstract

Automated task planning for robots, an important part of cognitive abilities of an autonomous agent, faces great challenges in that the sequences of events needed for a particular task are mostly required to be hard-coded. This can be a cumbersome process especially when the user wants a robot to learn a large number of similar tasks with different objects that are semantically related. Also, apart from learning distinct experiences correctly, significant or recurring experiences are expected to be learned better and be able to be retrieved easily relatively. In order to address these two challenges, we first propose a user preference-based dual-memory adaptive resonance theory network model (pDM-ART) which makes use of a user preference to encode episodic memories with various strengths

and to learn and forget at various rates. Over a period of time, as learned tasks are recalled, episodic memories undergo a consolidation-like process at a rate proportional to the user-preference at the time of encoding and the frequency of recall of a particular episode. These consolidated memories are stable and easier to retrieve. In the second part of the research, we propose a novel approach of user preference-based integrated multi-memory model (pMM-ART) that focuses on exploiting a semantic hierarchy of objects for enhancing the behavior of an autonomous agent in three ways. Firstly, the model is able to extend its knowledge of planning a task to an entire category of objects by learning to plan for only one object from the category. Not only is the extension possible to the category the object belongs to but also to other similar categories. Secondly, in certain situations, it is able to extend further by being able to plan on categories that have no objects planned on for. Lastly, it is able to recognize an erroneous category of objects that arrives in a retrieval cue. The memory model is capable of forming top down weighted connections between the concepts (consolidated episodes) and the semantic categories it has in its object fact map (OFM). Weighted connections (associations) also develop between semantic categories defining their distance relative to each other in semantic space based on the concepts and the attributes they share. We analyze both the models separately for a clearer understanding on how they play their role in trying to overcome the two aforementioned challenges. Simulation and experimental results are also presented.