My research interests lie in use of machine learning, intelligent agents and multi-agent systems to achieve autonomous optimization of large-scale heterogeneous systems. My work focuses on developing learning techniques for performance optimization, particularly in the presence of conflicting goals, as achieved by the adaptive collaboration between heterogeneous agents. I work on applying these techniques in smart cities domain, for example, Mobility as a Service, autonomous car sharing, intelligent decentralized urban traffic control and smart energy grid applications.
Below is the summary of my current and recent research projects.
It is estimated that autonomous cars will surpass manual forms of driving by 2040, thereby imposing themselves on both the built and the natural environment. Historically, changes in transport, such as the diffusion of the first automobiles in the 1920s, have led to dramatic changes in urban infrastructures, with severe negative environmental and social repercussions. By merging urban geography and computer science, Surpass will use Dublin as a case study to anticipate and map the infrastructural changes that autonomous cars will trigger, so that cities can use these changes as an opportunity to evolve in a sustainable way.
Given that self-driving cars are being diffused primarily through sharing services provided by international companies such as Uber, scholars argue that this new form of transport has the potential of reducing traffic congestion and car ownership, thereby opening up new opportunities to redesign cities. As having less traffic and cars in the city means that roads and parking spaces can be reduced, freed-up spaces could be reused for sustainability purposes and turned into, for example, gardens, bike lanes and renewable energy power stations. Through an innovative mixed-methods approach including foresight, scenario building, modelling and computer simulation, the Surpass team will produce the first study to estimate the impact that shared autonomous cars will have on urban traffic and car ownership, and quantify and map the urban space that can be freed-up.
As society demands more renewable forms of energy and the management of energy grids must take place in the face of the emergence of multiple micro-producers, there is a growing need for autonomic management of power distribution. LAMP applies agent learning and multi-agent techniques to the coordination of device usage in smart grids and optimization of renewable energy use. For more details see http://www.tcd.ie/futurecities/research/energy/.
REALT is a Real-time Adaptive Learning-based Traffic Control system for urban traffic optimisation. REALT is based on Distributed W-Learning (DWL) a multi-policy multi-agent RL-based optimization technique, enabling it to address multiple performance goals (e.g., overall traffic flow and special vehicle priority) simultaneously. REALT has been evaluated in simulations of Dublin and Cork city centres. For more details on the approach and results see Publications section.