Home // Traffic Simulacra: Generative Agents for Simulating Believable Mobility Behavior

Traffic Simulacra: Generative Agents for Simulating Believable Mobility Behavior

Type of thesis: Masterarbeit / location: Leipzig / Status of thesis: Theses in progress

The goal of this thesis is to improve current traffic simulations (e.g., the traffic simulator SUMO) by integrating generative agents, which are computational software agents that simulate believable human behavior using large language models. The goal is to make each agent (a person in the simulation) in SUMO behave more human-like, considering their opinions, experiences, and characteristics. For example, how would an agent decide whether to take the train or the car, when to drive to work, and how to react to a train strike?

The thesis will build on the idea developed in the paper Generative Agents: Interactive Simulacra of Human Behavior. This paper introduces an architecture that extends a large language model to store a complete record of the agent’s experience using natural language, synthesize these memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. The idea is to transfer these generative agents to the world of traffic simulation.

The expected results of the thesis are

A modified version of SUMO (or another simulator if better suited) that can incorporate generative agents as traffic participants.
A set of generative agents that can represent different types of human drivers and passengers in SUMO
A framework for evaluating the behavior of the generative agents in SUMO

Please contact Patrick Ebel if you are interested in this topic or if you require any further assistance. Below you can find some related work that might be of interest.

Related Literature: