Ronnie Smith (cohort student representative)

Research project title: 
Adaptation of Multi-Agent Systems in Ambient Assisted Living
Principal goal for project: 
To develop and evaluate methods of enabling Human-in-the-Loop (HITL) adaptation of smart home / Ambient Assisted Living (AAL) environments, where an ensemble of heterogeneous Internet-of-Things (IoT) devices, physically embodied (robotic) agents, and software-only agents form a Multi-Agent System (MAS). The overall intention is to reduce user burden in programming such systems by endowing the system with autonomous short- and long-term adaptation capabilities.
Research project: 

Personalisation and adaptation of smart home and Ambient Assisted Living (AAL) solutions is the subject of many existing works, which seek to embed and/or learn user preferences and needs. However, with a focus on lab-based evaluation, it has been easy to overlook the potential realities of what such systems might look like in the future: a range of heterogeneous platforms, devices, physically embodied, and software agents will be brought together to serve the specific needs of their users. These agents and devices together form an ecology, where the collective capability of members can be harnessed to achieve complex behaviour.

As such, it is important to consider scalability and interoperability in every aspect of future smart environments and AAL solutions. To avoid poor user acceptance, there must be mechanisms through which the aforementioned agents and devices can be autonomously configured and organised to provide the functionality that a user needs in a given scenario—with and without user input. Methods to enable this autonomous behaviour will need to consider the independence of agents as they pursue their own goals and tasks, while accommodating knowledge sharing and a form of centralised decision making.

Currently, my research is focused on the use of natural language in active learning scenarios in the home environment. In particular, I am considering a Human Activity Recognition (HAR) use case where pre-trained activity recognition models are updated to better suit a particular user and/or environment using labels gathered from the user via natural language. This requires that the system be able to initiate and maintain relevant contextual dialogues to extract the relevant information for inference of a valid existing (or in some cases, new) label.

View my publications: Google Scholar