A Cognitive Framework for Unifying Human and Artificial Intelligence in Transportation Systems Modeling

Abstract

Humans, as indispensable components in any transportation systems, have been very challenging to model and predict, especially in hypothetical scenarios. Adding further complexity is the increasingly important role of artificial intelligence and rapidly changing technologies and business models. We propose a modeling framework, CognAgent, which unifies the modeling approach of different types of autonomous entities from the perspective of cognition rather than revealed behaviors. This approach improves model flexibility, interpretability, and computational efficiency. Heterogeneous agents inherit from a single blueprint agent and interact with one another within the Physical Interaction module, the output of which is fed into the module of Space of Observables for agents to sense and perceive through noisy media of information transmission. Combining with prior knowledge, preprogrammed routines, emotions, and habits, agents make decisions on how to act in the Physical Interaction module. In CognAgent, information is a result of the change of perceived uncertainty, and therefore, consistent with the Information Theory. Owing to this explicitness of agents' cognition, the derived models become extendable to new technology and business models. Equity analysis related to cognitive limitations such as vision and hearing loss becomes also natural. The numerical example models explicitly humans and autonomous vehicles with heterogeneous information transmission, perception, and risk preference.