Investigating Requirements for Context-awareness and an Attempt to Implement it in Software Agents or Robots

Research on applying intelligent software agents and robots in everyday life is increasing. Their employment to take on dangerous tasks, intelligent agents as assistive systems for the elderly or as companions at work or at home to assist or facilitate the realization of certain tasks are of increasing importance in society today. Therefore, discussions arise about necessary prerequisites for software agents and robots to be able to solve these tasks, and whether a form of consciousness or awareness is one of these necessary prerequisites.

In this project, we dealt with this topic in a theoretical and an applicable way with the aim of implementing aspects of context awareness in a software agent or a robot. Philosophical, psychological, and neuroscientific aspects of consciousness as well as their modeling are a central topic of interest in many disciplines of cognitive science. The approach of this issue has the aim of assessing human and animal consciousness, better understanding its underlying mechanisms and to develop and implement computational models. Within the CHIST-ERA program 2010 ten aspects of "conscious systems or context-aware devices" were listed. The goal of the implementation realized within this project was to implement at least one of the listed aspects.

As a first step, we analyzed and reviewed existing literature for theoretical aspects of consciousness and context-awareness, discussed conceptual misunderstandings (e.g. regarding the definition of consciousness) as well as conceptual distinctions (e.g. access versus phenomenal consciousness). In the following, we concentrated on neural correlates and current empirical methods for measuring consciousness in the fields of psychology and neurosciences, e.g. oscillatory synchronization, lesion studies, illusions, priming, introspection, and imaging methods. The outcome of studies employing these methods influenced theoretical models of consciousness and context awareness we subsequently dealt with.

In order to equip complex systems as virtual agents and robots with aspects of consciousness, a cognitive architecture is needed that allows for these aspects. Within this project, we analyzed different cognitive architectures in detail to be able to select the most appropriate architecture for our implementation. Following this analysis, we judged the LIDA architecture to be the most appropriate architecture, as it allows for the most aspects of awareness. Although other architectures have a wider range of implemented cognitive mechanisms, most of these are specific to the area that they have been implemented for. The LIDA model, on the other hand, is more general and more plausible in its claims regarding consciousness and awareness, since it is based on the Global Workspace Theory of consciousness in neuroscience. Additionally, we decided to use the situational awareness model by Endsley as a theoretical basis.

As an implementation scenario, we decided the field of assistive systems, as there is growing need for assistive technologies to improve the quality of life of elderly or cognitively impaired adults. Over 11% of the world's population is over 60 years old, and this percentage is rapidly growing. Systems applied in this area have to be aware of external and internal situations and processes and act accordingly. Based on the LIDA cognitive architecture, we therefore developed CareBot, a cognitive software agent simulating a mobile assistive robot in a simple environment. The purpose of the agent is to perform fetch and carry tasks (a robot able to fetch food, drinks, or medicine for elderly or disabled individuals). The aim of this implementation was to demonstrate multiple aspects of context-awareness.

In addition to Endsleys situational awareness model we were able to implement five out of the ten aspects of context-awareness defined by the CHIST-ERA program. The CareBot agent possesses rich internal dynamics and is aware of the environment through multiple sensory modalities. The agent has a perception of its internal state. It is capable of autonomous construction of future plans and actions, as well as an analysis of the cost implications of such plans. By the integration of sensed external data and recall it is capable of generating an internal representation of itself and its environment. The agent is capable of autonomously and selectively directing its attention and computational processing power to address a current situation, external or internal to itself.

Following the CHIST-ERA 2010 definitions, the CareBot agent is context sensitive and possesses situational awareness according to Endsley's model. The presented work could be improved and taken further in two major directions. On the one hand, a major goal would be the implementation of the rest of the identified aspects of awareness in CareBot. Some of these aspects already have implementations in LIDA but have not been used in this simplified environment. On the other hand, implementing the CareBot agent on a real-world robot would be a challenging and demanding task and especially important in the field of assistive systems.

Research staff

• Robert Trappl
• Stephanie Schreitter
• Tamas Madl
• Dominik Schnitzer
• Karin Vorsteher

Sponsor