A Hierarchical Pedestrian Behavior Model to Generate Realistic Human Behavior in Traffic Simulation
DOI: 10.1109/iv51971.2022.9827035
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the critical need for realistic and controllable pedestrian behavior models in the development and testing of autonomous vehicles (AVs). Accurate simulation of vulnerable road users is essential for scenario-based testing, particularly for rare or dangerous interactions that may be underrepresented in real-world data. Existing microscopic models often lack extensibility or rely on non-deterministic "black-box" approaches that prevent engineers from explicitly injecting specific behaviors. To solve this, the authors present a hierarchical pedestrian behavior model that combines high-level decision-making via behavior trees with low-level motion planning using an adapted Social Force Model (SFM). This approach allows for explicit control over pedestrian actions while maintaining realistic physical interactions. The model architecture consists of three layers: a Behavior layer, a Maneuver layer, and a Motion Planner layer. The Behavior layer utilizes behavior trees to select maneuvers based on environmental states, such as traffic light colors or vehicle proximity. These trees are customizable, allowing testers to define specific decision logic, including varying levels of pedestrian aggressiveness. The selected maneuver is passed to the Maneuver layer, which translates it into low-level instructions comprising a waypoint, direction vector, and desired speed. Finally, the Motion Planner layer executes these instructions using the SFM, which calculates pedestrian motion based on attractive forces toward destinations and repulsive forces from other agents, vehicles, and boundaries. The implementation is integrated into the GeoScenario Server, enabling simulation of mixed traffic scenarios involving both vehicles and pedestrians. The model was evaluated using two naturalistic datasets collected in Ontario, Canada: one at a busy four-way intersection with signalized and unsignalized crosswalks, and another at a university location with a single unsignalized crosswalk. The evaluation methodology involved replacing individual empirical pedestrians in the recorded data with simulated agents initialized with only high-level routing information (start position, end position, and average speed). The simulated agents' trajectories and decisions were compared against the ground-truth data. To mitigate error accumulation in longer scenarios, the authors employed segmented scenario analysis. The results demonstrated that the model replicates real-world pedestrian trajectories with high fidelity, achieving an average deviation of 1.36 meters. Furthermore, the model achieved a decision-making accuracy of 98% or better, correctly replicating the high-level choices made by empirical pedestrians. The significance of this work lies in its ability to provide a highly controllable yet realistic simulation tool for AV testing. By decoupling decision logic from motion physics, the model allows engineers to safely generate and test critical scenarios, such as pedestrians running into traffic, without requiring manual trajectory adjustments. The explicit nature of the behavior trees facilitates the injection of specific behaviors, addressing a key limitation of existing models. This capability supports rigorous scenario-based testing, enabling the validation of AV responses to diverse and potentially dangerous pedestrian interactions in a safe simulation environment.
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | OpenAlex-citations | — | — | 1 | 2026-06-18 |
| archive | success | unpaywall | — | — | 2 | 2026-06-25 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-18 |
| chunk | success | chunk | — | — | 1 | 2026-06-18 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-18 |
| promote | success | — | — | — | 1 | 2026-06-18 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-18 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Theoretical Contribution: computational model