Memory modules for a cognitive architecture to execute complex manipulation tasks for Personal Service Robots

Abstract

This work presents an approach to improving the autonomy of robots. This is done by adding cognitive features with the aim of allowing these robots to be used in homes or care facilities with little supervision. Specifically, my approach is to build knowledge-processing memory modules capable of acquiring, retrieving, storing and updating knowledge. Using modules that simulate memory allows robots to process knowledge and perform better in human environments. In particular, this work focuses on Personal Service Robots executing complex manipulation actions embedded into everyday tasks. I aim to use techniques of learning from human examples to enable robots to perform everyday tasks in household environments. A human activity recognition system for demonstrations from a Virtual Reality environment was implemented to record and analyze those examples. This system is capable of segmenting trajectories and events into actions and sub-actions. These are stored in a hierarchical data structure called episodes, used by the memory modules to give the robot insight into how to execute such actions. Also, this research developed knowledge-processing memory modules to extend a cognitive architecture. These memory modules are part of Long Term and Short Term Memory. This architecture is transferable to diverse robotic platforms. It was tested in five robots from different vendors and body configurations. This shows that the knowledge can be used depending on the robot’s necessities and capabilities. The memory modules allow the robot to decide whether and how to adapt actions. This way, the robots can perform complex manipulation actions similar to humans. The experiments in this work were performed in a simulated kitchen environment. There, the five robots performed actions embedded in the cooking tasks. The results show an improvement in the action execution success by eleven percent and a similar percentage reduction in the time required to perform the tasks. This work draws inspiration from linguistics, psychology, neuroscience, cognitive science and computer science to create Knowledge Bases. They contain concepts from natural sciences, actions, skills, objects, robotic platform capabilities and experience.