TravelAgent: Generative Agents in the Built Environment

Understanding human behavior in built environments is critical for designing functional, user-centered urban spaces. Traditional approaches—such as manual observations, surveys, and simplified simulations—often fail to capture the complexity and dynamics of real-world behavior.

To address these limitations, we introduce TravelAgent, a novel simulation platform that models pedestrian navigation and activity patterns across diverse indoor and outdoor environments under varying contextual and environmental conditions. TravelAgent leverages generative agents integrated into 3D virtual environments, enabling agents to process multimodal sensory inputs and exhibit human-like decision-making, behavior, and adaptation.

Through experiments, including navigation, wayfinding, and free exploration, we analyze data from 100 simulations comprising 1,898 agent steps across diverse spatial layouts and agent archetypes, achieving an overall task completion rate of 76%.

Using spatial, linguistic, and sentiment analyses, we show how agents perceive, adapt to, or struggle with their surroundings and assigned tasks. Our findings highlight the potential of TravelAgent as a tool for urban design, spatial cognition research, and agent-based modeling. We discuss key challenges and opportunities in deploying generative agents for the evaluation and refinement of spatial designs, proposing TravelAgent as a new paradigm for simulating and understanding human experiences in built environments.

A schematic representation of the TravelAgent system. (left) TAs are initialized with various parameters defining the agent persona and its environment. (center) At each step, the agent employs Chain-of-Thought (CoT) to process sensory inputs, plan actions, and make decisions. (right) The agent executes actions within the environment and updates its internal memory based on its experiences.

Scenario Descriptions

Scenario 1 - Base: A bright summer morning in Kendall Square, Cambridge, MA. Modern glass buildings and bustling streets with pedestrians, cyclists, and outdoor cafés creating a lively and vibrant atmosphere.
Scenario 2 - Winter: A snowy winter morning in Kendall Square, Cambridge, MA. Heavy snow blankets the streets and modern buildings, with bundled-up pedestrians, and a quiet stillness filling the air.
Scenario 3 - Tokyo: A bright summer morning in Tokyo, Japan. Mix of traditional wooden buildings and modern structures. Bustling streets filled with pedestrians, cyclists, and outdoor tea houses creating a vibrant atmosphere.
Scenario 4 - Night: A quiet summer nighttime streetscape in Kendall Square, Cambridge, MA. Streets are softly illuminated by warm streetlights and the gentle glow of modern office buildings with large glass facades.
Scenario 5 - Persona: The agent's persona was varied between a 30-year-old female and a male around the same age. This was meant to examine whether the agent's gender influenced its decision-making process.

Scenario Matrix

Table 1. Associations between Season, Location, Time, Persona, and Scene
Season	Location	Time	Persona	Scene
Summer	Kendall Square, Boston	Morning	30-year-old female/male	Scene 1
Winter	Kendall Square, Boston	Morning	30-year-old female/male	Scene 2
Summer	Tokyo	Morning	30-year-old female/male	Scene 3
Summer	Kendall Square, Boston	Night	30-year-old female/male	Scene 4

Scene 1 - Persona

Scene 3 - Location

Spatial Analysis of `Train Station' Experiment. For each scenario, we evaluate the agent's successful and failed paths. If the agent reaches and recognizes the subway station by declaring `stop', the path is considered successful, and the agent is given a new subtask. The top left figure show both a successful and a failed path in the `Base' scenario. The top row in the bottom figure shows the paths of all agents in all other scenarios, and the bottom row shows the spatial aggregation of decision points across all scenarios. Notably, the agents in the `Night' scenario have the most failed and inconsistent paths, which is also reflected in more sparse decision point area. Conversely, the agents in the `Winter' scenario have more consistent paths, with a clear aggregation of decision points early in the simulation.

Analysis of the 200 most frequent words in the agents' planning streams, clustered by scenario. The dominant terms correspond to the agents' tasks, goals, actions, and interactions with their immediate surroundings. Across all scenarios, the most prevalent terms are related to the agents' target and goal, such as `station', `target', and `move'. The agents also reflect on the shape, form, and materiality of their environment, with terms like `bricks', `glass', `flat', and `building' appearing frequently.

Sentiment analysis of the agent's `thoughts' and `observations' streams. Each experiment is displayed as a path with the agent's decision points. The distance between each decision point refers to the distance traveled in `Forward' actions. Sentiments are classified into three categories: positive (blue), neutral (gray), and negative (light pink) using NLTK’s VADER~\citep{hutto2014vader}. In successful experiments, the red dots represent the "Finish" steps, where the agent is given an additional randomized subtask. Notably, in failed paths, there are clear clusters of `search' actions, in which the agent aims to reorient itself, usually leading to a negative sentiment. In successful paths, the agent's sentiment is more positive, and search steps are less frequent. In sub-tasks, mixed sentiments appear, alongside some `search' clusters.

BibTeX

@article{noyman2024travelagent,
      title={TravelAgent: Generative Agents in the Built Environment},
      author={Noyman, Ariel and Hu, Kai and Larson, Kent},
      year={2024},
      journal={arXiv preprint},
    }

TravelAgent: Generative Agents in the Built Environment

TravelAgent simulates human-like behaviors and experiences in diverse built environments

Abstract

TravelAgent Scheme

Scenarios

Scenario Descriptions

Scenario Matrix

Scene 1 - Persona

Scene 3 - Location

Scene 2 - Season

Scene 4 - Time

Evaluation

Spatial Analysis

Term Frequency Analysis

Topical Modeling

Related Works

BibTeX