(PI UMa: Nuno Nunes, Co-PIs CMU: Lucio Soibelman /H.Scott Mathews)

Starting Date: 1 Jan. 2009, Duration: 36 months, Person * month in this task: 51

Results

The objective of this work package is to explore the use of multi-sensor systems in conjunction with contextual inquiry HCI techniques to detect and understand significant human activities related to resource consumption in a domestic environment. Through combining sensing systems with exploratory HCI techniques we aim to achieve a deeper understanding of the human behavior related to resource consumption in the home, thereby providing evidence that could be used to motivate sustainable practices (WP1) or design innovative services (WP4). The output of this work package will be a robust and tested low-cost sensing infrastructure with associated machine learning systems informed through in-depth contextual inquiry that can be used to develop compelling applications of sensor-based statistical models.

This work package will explore the infrastructure in an innovative way:

· Sensing Infrastructure: At the technological level we will focus on monitoring energy and water consumption through low-cost non-intrusive sensing systems that can detect significant human-activities through machine learning systems, which can classify and act on sensed data such as detecting routines in people’s behaviors. This work package will look at processes for simplifying complex multi-sensor systems resulting in cheaper and easier systems to deploy, and software infrastructures for enabling the collection, inference on, and distribution of sensed data;

· Development and Evaluation of Context-aware Applications: Context-aware computing enables applications that sense the environment, model situations and act appropriately. This work package provides a unique opportunity to test innovative context-aware applications that take advantage of the sensing infrastructure to explore the design space surrounding sensor-based statistical models of human situations at the home. This could lead to the development of new toolkits that facilitate development and deployment of context aware applications that could go beyond the domain of sustainability;

· Visualization and Feedback: in conjunction with WP 1, the information gathered in this task will be used to create innovative visualization techniques that will close the feedback cycle of providing consumers with relevant and in-context information about home consumption usage. These visualization techniques will be evaluated systematically to maximize conservation opportunities and motivate behavior change.

The results of this work package will be a range of low-cost, easy to deploy, technological solutions to unobtrusively motivate sustainable practices and provide design opportunities for innovative services in the broader context of the Madeira Living Lab - meaning that technologies are required to be robust and appealing as innovative commercial solutions for the companies involved and also practical and accessible so that user-centric evaluation and feedback can effectively contribute to the open innovation process.

Description

Most of the groundbreaking work in human assistance and context-aware applications was traditionally applied in small pilot projects in the areas of defense and healthcare. In this work package we propose to apply several methods and techniques to the domain of sustainability in the broader context of the Madeira Living Lab, posing several interesting research and technical challenges, such as the implications of widespread deployment and the broader diversity of the user population.

Successfully deploying sensor-based statistical models is currently difficult. We will look into low cost non-intrusive sensor technology that could be used to unobtrusively understand human situations that are inherently ambiguous. The problem is typically not that models are incapable of learning about human situations, as the machine learning community has developed powerful methods for learning from labeled datasets. Still, the use of machine learning generally requires significant specialized knowledge, particularly with regard to bridging the gap between sensed context and high-level features appropriate for use in machine learning algorithms.

This work package will focus on residential consumer behavior through monitoring of energy and water consumption patterns providing homeowners with detailed information about their operating schedule. Our approach will not target efficiency-motivated equipment upgrades. Rather, it will build on existing research from groups at CMU that have explored practical and technically feasible solutions for monitoring electricity and water consumption through inexpensive non-intrusive techniques that extract detailed information from the aggregate source. This approach contrasts with traditional techniques that involve installing sensors in individual apparatus, thus making the technology expensive, intrusive and difficult to manage.

Although the sensing technology plays an important role in capturing detailed usage data from consumers this project aims at understanding behavior patterns and creating opportunities to change that behavior. Thus this work package will also involve the application of HCI techniques to understand and collect additional information from consumers in context. These techniques will be used to complement the automatic sensed data and provide increased understanding of the context in the home.

Task activities:

· Recruitment of a significant sample of household consumers in the context of the Madeira Living Lab;

· Installation of DAQ and other infrastructure for aggregate sensing equipment for the sample study;

· Conduct user studies, including contextual inquiry and ethnographic research;

· Installation of the data collection and feedback infrastructure;

· Prototype visualizations and the feedback structure to analyze behavior change;

· Prototype evaluation and re-design in conjunction with WP1 and 4;