This is the major research thrust in the CoSMOS lab with multiple projects funded by NSF, USDOT, Tennessee Department of Transportation (TDOT), Department of Energy (DOE), Oak Ridge National Lab and Volkswagen research. In this research domain, we investigate methods of optimized coordination among different elements of the traffic network, such as CAVs, platoons, traffic signals, etc.
We are investigating the synthetic-to-real domain adaptation problem using a custom-built variant of the Cycle-GAN network. In collaboration with ORNL, we are also developing a state-of-the-art vehicle-in-the-loop simulator, where a real vehicle can be mounted on a set of steerable hub-dynamometers which allows complete hardware in the loop testing of vehicle response to (synthetic) stress cases.
Oak Ridge National Laboratory – Regional Mobility Project: In this project, in collaboration with ORNL and the city of Chattanooga, we are utilizing high resolution real-time traffic flow data from Gridsmart cameras for decentralized optimization of signal phase and timing of participating intersections for increased throughput and decreased delays.
NSF-CIS
Human drivers in legacy vehicles, pedestrians and cyclists represent stochastic disturbances in an otherwise controllable system. As part of a completed NSF CIS project, my research has focused on the development of a new negotiation-based multi-agent reinforcement learning (MARL) framework capable of reaching a correlative solution among a system of multiple agents (human drivers and computer-drivers) through a model-free iteration of action forecasts. We have deployed the framework in a variety of traffic situations (such as on and off ramps, lane closures, platooning vehicles, etc.) to optimize traffic flow (measured by increased throughput, trajectory smoothness, fuel-efficiency, etc.) on links and at nodes.
We are investigating the effect of information fusion from multiple data streams – namely, the driver, vehicle kinematics and the surrounding environment to classify distracted driving from nominal driving – the goal here is to design a more intelligent driver assist system.
Since multi-agent interaction is the main focus in the CoSMOS lab, we have also acquired a significant amount of research expertise in modeling and optimizing crowd behavior, specifically in emergency evacuation situations. Herding characteristics can be observed during emergency evacuation from buildings under threat, where, in choosing escape routes, homophily often trumps judgement. In fact, the two dynamics of movement and decision end up intertwined – clusters form due to movement and movement is affected by the cluster. My group has led pioneering work [4, 5] in analyzing this coupled dynamical system and investigated the effect of leader following, route familiarity, panic, impatience and herding on the overall egress efficiency using language-theoretic discrete choice models.
Our group found that factors such as how receptive the crowd is to opinion sharing, and how fast the individuals tend to change their exit choice when confronted with crowded lanes/bottlenecks, affect the crowd’s evacuation time. Ideally, a tolerant rational crowd with well-informed leaders/strong opinion holders is well-suited for a quick evacuation of a building. Herding is not detrimental for evacuation. However, over-herding can lead to under-utilization of all the available routes and an increase in the evacuation time.