Skip to main content

MODULAR HYBRID ARTEFACTS WITH ADAPTIVE FUNCTIONALITY

Objective

The ORESTEIA project tries to create a guidance system for humans, for more efficient and less hazardous living and interacting with their environment, through a set of decision-making facilities embedded in the environment and suitably adapted to the particular user. Inputs arise at low-level from a range of classes: physiological, environmental, and other such systems. There is a level of known expertise to be included, as well as some form of combination of sub-symbolic with symbolic processing in the guidance system. A multi-level attention-based agent approach will be developed to solve a range of decision/guidance problems that arise in the Disappearing Computer (DC) initiative. The project considers both development of single agent architecture to handle such problems, and also how multi-agent systems can be constructed by suitably extending such single agent systems.
ORESTEIA also aims at investigating enabling technologies for DC in the form of energy harvesting and low power wireless communications.
The ORESTEIA project tries to create a guidance system for humans, for more efficient and less hazardous living and interacting with their environment, through a set of decision-making facilities embedded in the environment and suitably adapted to the particular user. Inputs arise at low-level from a range of classes: physiological, environmental, and other such systems. There is a level of known expertise to be included, as well as some form of combination of sub-symbolic with symbolic processing in the guidance system. A multi-level attention-based agent approach will be developed to solve a range of decision/guidance problems that arise in the Disappearing Computer (DC) initiative. The project considers both development of single agent architecture to handle such problems, and also how multi-agent systems can be constructed by suitably extending such single agent systems.
ORESTEIA also aims at investigating enabling technologies for DC in the form of energy harvesting and low power wireless communications.

OBJECTIVES
- Create a general world representation by integrating multiple sensor data (Data fusion);
- Create a distributed, hierarchical, attention control based, goals-oriented decision making architecture (Attentional Agents);
- Support emergent behaviour in a collection of artefacts (Emergence);
- Create a system, which can adapt to user changing needs over time and learn using multiple paradigms including symbolic and sub-symbolic approaches. The system uses context information and defines the concept of the User Profile (Adaptivity);
- Research on enabling DC technologies in the form of micro-power generation and low power radio communications (Power generation).

DESCRIPTION OF WORK
The main project issue is to construct a distributed, highly robust, effective decision making system for user hazard avoidance and well being. To achieve such a system one has to face several important aspects of the problem. One needs to use sensor data coming from the user(s) and the environment for observing rather than asking constantly the user(s). This data must be interpreted in a specific way taking further input from contextual information so as the user intention can be made clear. Multi-sensory data integration and the related problem of state representations are very hard in the general case. The ORESTEIA system divides the input space in a number of modalities for achieving local decision-making per modality and then a combined overall decision is reached through multi-modal integration. The problems of state evaluation (local and global) and control are in the heart of any guidance system. ORESTEIA uses attention control as a way to prioritise responses that result from state evaluations. The principal idea is that not all events in the environment of the user are equally important, so priority is given to unexpected and unidentified cases, which potentially carry the highest risk for the user.
Finally, supporting emergent behaviour becomes progressively harder when we move from a compositional approach to evolution through gathered experience to ad-hoc (in time and space) formed artefact collections. To this end ORESTEIA devised a computational model that is goal-oriented and that constrains the system actions so as to guarantee the highest survivability, the fast system response and the handling of conflicting goals.
The ORESTEIA project has been actually made up of two components: 1) that for the first year, up to the date of the first review (Feb, 2002), and 2) that developed from the requirements of that review, that there be a change of program and of view.

The first component focused on the creation of artefacts responsible for interaction with humans, possessing modularity, hybrid architecture composed of subsymbolic and symbolic components which continuously interact with each other with adaptable functionality. Of particular importance to the project has been the design of collections of artefacts providing health status analysis and hazard avoidance. To achieve this objective, the ORESTEIA project focused on the creation of artefacts which include intelligence, in the form of a-priori knowledge/rules, related to their functionality, adapt their performance to the specific environment they are used into, detect whether they need to refine their a-priori knowledge according to the data they receive from the environment and in such a case implement this refinement or change. The project in its first component defined, designed and implemented a prototype of a generic hybrid architecture, which can form the core of such intelligent artefacts. This architecture is composed of a subsymbolic and a symbolic part, which interact with each other during training, operation and adaptation. The functionality of the artefacts takes advantage of the principled interaction between the symbolic and subsymbolic components; this was based on a top down procedure in which subsymbolic components supplement and link rule based components in a compositional way, and a bottom up procedure exploiting the ability of well assessed configured to generate symbolic rules within suitable confidence levels. The Probably Approximatevely Correct (PAC) learning paradigm for Boolean functions and fuzzy inference machines have been used to build symbolic components capable of interacting with the subsymbolic ones. Particular importance was given to the capability of the artefact to adapt to the actual environment that it is used into.

The milestones of this component have been : (1) Development of the generic hybrid symbolic-subsymbolic artefact, (2) On-line adaptation of the artefact's behavior, (3) Implementation and Validation of architecture of the artefact and of its functionality, (4) Standardization of exchanged data (signals, features, decisions) format and protocols, (6) Implementation and validation of appropriate micro-power generation modules, (7) Generation of a collection of artefacts for user health status analysis and hazard avoidance. The second component of ORESTEIA focused on the development of multi-level attention-based agents adapted to solve a range of decision/guidance problems that arise in the Disappearing Computer framework. The problems arise from sensors of various types, some worn by humans, others in devices, such as cars, being used by the humans. The project considered both the development of the architecture for a single agent to handle such problems, and also how multi-agent systems can be constructed by suitably extending such single agent systems. Finally the project investigated enabling technologies for the Disappearing Computer in the form of energy harvesting and low power wireless communications, giving important constraints on the abilities of the agent systems to be developed. The emphasis has been on the high level problems of integration and attention . The approach was based on techniques for hierarchical attention processing with feedback of solely sub-symbolic, or joint sub-symbolic/symbolic information, with additional analysis of medical data.

The resulting system is highly autonomous, responsive and robust, acting on behalf of the user for: i. Managing repetitive and trivial jobs, ii. Provide indication of abnormal user activity and state, iii. Provide planning facilities, iv. Provide information filtering facilities, v. Maintain good user state (physiological, psychological, etc). This has been achieved through accomplishing the following partial developments: Creation of scenarios with data collected and used in testing of the attention-based agent guidance systems. Preprocessing the data, so as to remove noise or outliers, and avoid simple mistakes creeping in at early stage. Design of attention-based artefacts with embedded guidance powers and inherent adaptability in a given modality. Design of multi-modal Attention Agents. Design of collections of collaborating artefacts and definition of roles within the collection. Design of micropower generation modules for artefact supply. Prototyping and implementing specific artefacts' collections for user health status analysis and hazard avoidance.

Funding Scheme

CSC - Cost-sharing contracts

Coordinator

ALTEC INFORMATION AND COMMUNICATION SYSTEMS S.A.
Address
Patmou 12
15123 Maroussi - Athens
Greece

Participants (5)

IMPERIAL COLLEGE OF SCIENCE, TECHNOLOGY AND MEDICINE
United Kingdom
Address
South Kensington Campus
SW7 2AZ London
INSTITUTE OF COMMUNICATION AND COMPUTER SYSTEMS
Greece
Address
Patission Street 42
10682 Athens
KING'S COLLEGE LONDON
United Kingdom
Address
Strand
WC2R 2LS London
STMICROELECTRONICS S.R.L.
Italy
Address
Via Olivetti 2
20041 Agrate Brianza
UNIVERSITA DEGLI STUDI DI MILANO
Italy
Address
Via Festa Del Perdono 7
20122 Milano