Dr Enrico Gerding is an Associate Professor in the Agents, Interaction and Complexity (AIC) research group in the School of Electronics and Computer Science (ECS) at the University of Southampton. His main area of research is autonomous agents and multi-agent-systems, and the intersection with other fields such as economics, human-computer interaction and machine learning. In particular, he is interested in market mechanisms and automated negotiations, and how incentives can be designed in such a way to make systems more efficient. He has over 100 peer-reviewed publications on agents and multi-agent systems, and has published in the main AI conferences journals including AAMAS, IJCAI, AIJ and JAIR. Example topics of interest include: mechanism design, and its application to energy markets and the development of community energy markets; privacy and consent and the use negotiation for automating consent decisions; smart mobility and mobility as a service; air pollution monitoring and the use of crowd sourcing. Within the AutoTrust project, he leads the Automation theme.
Automation Theme: Within the IoV a user can easily become overwhelmed by the number of decisions that need to be made, from whether or not to join a platoon to consenting whether certain data or private information is sent to other vehicles or other systems. These decisions can be complex, involve other stakeholders, and several trade offs need to be considered, e.g. time vs effort. This theme investigates ways to support and even automate such decisions within a range of IoV scenarios. This involves several key challenges: (1) how to model the user preferences and infer user preferences from limited knowledge or peer knowledge, (2) how to automate the negotiation between multiple stakeholders when there is a conflict of interest (e.g. who should be at the front in the platoon, or who should get priority at intersections, or what data should be shared), (3) how to reward users and incentivise the “right” type of behaviour, (4) how make optimal decisions (e.g. in the context of routing) in the face of uncertainty.
Title: Intelligent Traffic Control to Improve Congestion. How Does the Amount of Participation Affect Control Effectiveness?
Traffic intersection control has clear potential of improving efficiency from the Internet-of-Vehicles: by communicating vehicle location, speed and even the entire planned trajectory to the infrastructure, traffic lights can anticipate arriving vehicles and plan the traffic light control in an intelligent manner. This problem quickly becomes very complex when considering multiple intersections and coordination between them. However, even for a single intersection, there is a question about how much information is needed to make significant efficiency gain. In particular, we envision a future with a combination of different types of vehicles: some communicate their information to the system, but others do not for a variety of reasons (e.g. privacy or effort or lack of technology). Given this, the specific questions we pose in this project are: First, is there a participation level above which there is no more benefit? Second, if a certain percentage of others already participate, is there still an incentive for any individual to participate or can they simply free ride; and what is the threshold of participation where this free-riding may become an issue?
Southampton undergraduate student Adam Barcock has investigated these questions during his summer internships. In particular, he has implemented a well known optimization algorithm for controlling the phases of the traffic lights at an intersection based on dynamic programming, and integrated this algorithm into the realistic SUMO traffic simulation platform. He has modified the algorithm to deal with partial participation of vehicles, and is measuring the travel duration and waiting as a way to determine the efficiency of the system as a whole, as well as the individual benefit from participating. Initial results are interesting and, to some extent, counter to what we expected. As expected, the system efficiency increases as a whole as more people participate. Interestingly, the individual benefit remains significant even when a large proportion of the population is already sending their data to the system. Further work is needed to see what happens if the algorithm is expanded to uses machine learning to predict future arrivals.