AGE-P: The Appliances-Go-Evolution Platform

Keywords: Smart Appliances Ensemble, Self-Organization, Artificial Evolution Office Lighting Scenario, Variable-Size Genotypes, CC1010 Wireless Sensor Nodes

Motivation: The term Smart appliances refers to everyday-life devices that do something useful and that are equipped with some computational resources. Then these appliances form an ensemble, if they coherently cooperate in order to support the its users in an non-intrusive way. These smart appliances ensembles can characterized by the following properties: (1) devices join and leave with out notice, (2) therefore, no instance is aware of the current number of participating devices, and (3) the devices are supposed to cooperate without any human intervention and/or explicit configuration. Because of the uncertain number of ensemble members, classical optimization procedures, such as gradient descent, evolution strategies, and genetic algorithms, cannot be used, since the term ’object’ is useless in its classical sense. In addition to the obvious technical issues, the ensemble has to be able to somehow derive the users’ intentions, in order to cooperate in a useful way.

Goals: The main goal of the project is to develop a self-organization scheme that accounts for the specifics of the problem at hand. This new scheme should have the following properties:
(1) the participants may not be aware of the others,
(2) the devices should employ a low-cost wireless communication module,
(3) the appliances should be able to form an ensemble even if they do not know of each other and even if they do not share a common ontology, and
(4) communication and feedback should be based on physical quantities.

Approaches: This project explores a combination of the following approaches:
(1) The self-organization process is based on the evolution strategy, a member of the class ob optimization procedures, also known as evolutionary algorithms.
(2) A typical setup consists of several sensors and several actors, which all measure and contribute the very same physical quantity; further physical quantities might also be present.
(3) The evolutionary algorithm does not maintain an object in the classical sense as it represents a particular configuration; rather, the genotype is distributed among all the actors.
(4) Fitness evaluation is not done by a central entity; rather, all sensors broadcast their differences between their sensor readings and the target values.
(5) The sensors’ target values are specified by the user and/or a system wide intention recognition module, which is part of a smart appliances environment. (6) The concepts are evaluated in a typical large office space lighting scenario, which consists of various light sources, sensors (e.g., on the users’ desks), window blinds, and the sun.

Results: For the office lighting scenario, the team is developed a prototype, which consists of:
(1) off-the-shelf torchiere with an attached, custom-made wireless communication controller (CC1010),
(2) light sensors with the very same wireless controller, and
(3) a test and debugging module (a regular PC).
This prototype has been working for about three years in our office.

Research Team: Stefan Goldmann, Matthias Hinkfoth, and Ralf Salomon

Contact: Ralf Salomon

Selected Publications:

• Goldmann, S. and Salomon, R., RfM: A Radio Flashing Tool for Microcontrollers.
• Salomon, R. and Goldmann, S. AGE-P: An Evolutionary Platform for the Self-Organization of Smart-Appliance Ensembles. In Raymond Chiong (ed.): Nature-Inspired Informatics for Intelligent Applications and Knowledge Discovery: Implications in Business, Science and Engineering, IGI Global, pp. 182-203, 2009.

Partners: