Abstract— CSIRO is using wireless sensor network technology to deploy “tiny agents”, working as autonomous controllers for individual pieces of electrical load/generation equipment in a distributed energy system. The tiny agent concept is a novel application of wireless sensor networks, providing the benefits of multi-agent systems science in a cheap, mobile, and highly distributable platform. However, the performance constraints inherent to wireless sensor networks mean the real-world realization of a tiny agent system is a significant challenge. This article details our work on tiny agents. We include a brief review of multi- agent system benefits, and then discuss the challenges inherent to the tiny agent concept. We also detail our applications work in applying wireless sensor network technology to operate as tiny agents, with a focus on intelligent heating, ventilation and air-conditioning control. Index Terms— Demand Response, Distributed Energy, Multi-Agent Systems, Wireless Sensor Networks I. INTRODUCTION SIRO has had significant experience applying wireless sensor network (WSN) technology in real- world deployments as part of our work with distributed energy, an exciting field of research designed to change the way electricity networks operate. Until recently, we have used WSN hardware in the relatively traditional role of distributed sensors. We are now working on a novel application of WSNs as agents for the intelligent control of energy resources, including heating, ventilation, and air-conditioning (HVAC) systems. CSIRO is working on the decentralized control of groups of energy resources, as we believe such an approach has significant advantages over the more traditional centralized control systems currently used in electricity networks [1]. In particular, our technology revolves around the application of multi-agent systems research, from disciplines including machine learning and complex systems science, to the management of large energy devices. In essence, entities in the electricity network (such as HVAC systems or small electricity generators) are represented and managed by agents, where an agent gathers data, makes decisions, and takes actions on behalf of a certain entity. A collection of agents is able to achieve quite complex outcomes, often displaying behaviour that would not be possible from one single agent- a system-wide behaviour emerges from the network of individual devices. The agent concept in general, and the science of multi-agent systems in particular, is a very complex and diverse field. Entire books have been dedicated to the area of multi-agent systems science, and the reader is directed to examples such as [2] for a more comprehensive review of this area. In using multi-agent systems techniques for managing energy resources, our general approach is to deploy an agent for each individual (significant) load or generator in the electricity network. In a domestic situation, individual agents may be associated with devices such as an air- conditioner, hot water heater, washing machine or dishwasher; and these agents can then collaborate to minimize the energy consumption of a house. Basic requirements of an agent, physically located at a distributed energy device, are a method for interacting with and controlling the device, a computing ability enabling the agent to manage the requirements of the local device, and communications ability so the agent can interact with other agents in the multi-agent system. Until recently, our applications of WSN hardware have been relatively traditional- sending data back to a distributed energy agent for collation and further processing. For example, we often use WSNs for integrating electricity demand management techniques with HVAC systems, where WSNs can be used to provide high-granularity sensor data such as temperature and occupancy; are relatively cheap, and easily deployed. Having already realized the benefits of a decentralized approach to controlling significant electricity loads and generators, we are now working on further decentralizing our management and control system, pushing the application of agents to lower levels of the electricity network. We are working on the deployment of tiny agents, where the functionality of wireless sensor network hardware is extended from a sensor-centric behaviour, to implementing the role of distributed, intelligent control agents. In this concept, a wireless sensor based tiny agent is responsible for controlling one particular, discrete device, based on inputs from the surrounding environment. Example applications of a tiny agent may include: The Tiny Agent- Wireless Sensor Networks Controlling Energy Resources Glenn Platt 1 , Joshua Wall 1 , Philip Valencia 2 , John K. Ward 1 1 CSIRO Division of Energy Technology, PO Box 330, Newcastle NSW 2300, Australia 2 CSIRO ICT Autonomous Systems Laboratory, PO Box 883, Kenmore QLD 4069, Australia Email: {glenn.platt, josh.wall, philip.valencia, john.k.ward}@csiro.au C 42 JOURNAL OF NETWORKS, VOL. 3, NO. 4, APRIL 2008 © 2008 ACADEMY PUBLISHER