When a hurricane or tunnel collapse hits, people flee based on gut instinct, not rational calculation—and current evacuation models fail to account for that. NEXUS is building a new generation of travel behaviour models that fuse passively collected mobility data (from smartphones, smart cards, and social media) with brain-signal measurements from wristbands and responses from virtual-reality simulations of wildfires, storms, or collapsed tunnels. The core problem is that existing emergency models assume people make logical choices under stress, ignore psychological factors like risk-taking propensity and collective behaviour, and rely on small surveys where people poorly recall their past actions. By combining choice modelling, ubiquitous computing, and machine learning, the team aims to produce models that realistically predict whether someone will evacuate, when, and by which route. If successful, planners and policymakers could test the impact of different emergency scenarios—such as road closures or staggered evacuation orders—far more robustly, potentially saving lives during natural disasters and transport network disruptions.
View original technical description
The mobility landscape is undergoing rapid changes with the advent of new technologies and the growing complexities of travel patterns. NEXUS focuses on developing next-generation mathematical models of travel behaviour that can better predict the activity and travel decisions in this changing landscape. This is being achieved by developing new frameworks that bring together Choice Modelling (CM), Ubiquitous computing (UC) and Machine Learning (ML) techniques to utilise passively generated real-world mobility traces (from public transport smart cards, mobile phones, etc.), neurophysiological signals and virtual reality (VR) to model decision making in future scenarios. The research focus for the 1st phase has been to model day-to-day activity and travel decisions (choice of travel mode, destination, time-of-travel, etc.) - in the context of current and emerging modes (self-driving cars, air taxis, hyperloops) and in regular and challenging environments (e.g. pandemic, economic and social unrests). In the next phase, I plan to focus on extending the theme of fusing different types of data and bridging CM, UC and ML in the context of emergency situations - during natural disasters and acute transport network disruptions, in particular. The limitations of the current behaviour models for emergency situations (e.g., whether or not to evacuate, when to depart, which mode and route to take, etc.) arise from multiple factors. Firstly, they assume travel choices in such situations are based on rational decision-making principles which is often not the case. Rather, the choice alternatives in such scenarios have varying levels of uncertainty and the decisions very often are impulsive and based on 'gut feeling'. Secondly, the current models do not account for the myriad of psychological factors that could influence an individual's decision in such difficult situations, for example, the risk-taking propensity, the perceived effect of stress or the thinking process in general. Thirdly, very often there is an element of 'collective behaviour' in such situations where a group of decision-makers influence each other's decisions consciously or unconsciously. This is challenging to capture in the modelling framework. Finally, the data used for developing the models rely on small-scale surveys where travellers are asked to report/log their past behaviour or to state their choices based on descriptions of hypothetical scenarios.These very often are not reliable measures of real-world travel behaviour. On the other hand, advanced technologies and machine learning methods have made it possible to measure the 'cognitive state' of travellers with simple wristbands, discreet clip-ons and smartphone-based sensors and infer their 'thinking process' from brain images. Further, advances in virtual reality (VR) technology have made it possible to immerse travellers in simulated emergency scenarios (e.g. VR replications of wildfire, stormy weather preceding hurricanes, collapsed tunnels, etc.) and obtain more realistic responses. These advances hold promise for producing more realistic inputs leading to a step change in modelling travel behaviour in emergency situations. The focus of the renewal phase of NEXUS will be to address the research gaps in modelling travel behaviour for emergency situations by leveraging novel forms of data. These will include (a) real-world mobility data generated from mobile phones, social media and other passive sources; (b) experimental data on travel behaviour from VR settings of hypothetical emergency scenarios. The applicability of the framework we are developing in the 1st phase of NEXUS to unify CM, UC and ML will be extended to achieve the objectives. In addition, other approaches like 'simulation-based data fusion' will be explored and compared. The developed models will enable planners and policymakers to test the impact of alternative scenarios in emergency situations in a more robust manner.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know