Source-linked AI summary
Pedestrian Archetypes Extension -- More Pedestrian Models for Autonomous Vehicle Safety Testing
Taorui Huang, Namita Gaidhani, Ritvik Bansal, S M Jubaer, Regina Lim, Rhett Zhao, Gavin Rafael Selin, Sunnie Deng Gao, Hasnain N Syed
TL;DR
Autonomous-vehicle safety models often represent pedestrians through isolated actions, leaving recurring and difficult-to-predict behavioral patterns incompletely modeled. This preprint extends a 12-archetype taxonomy with seven additional archetypes identified through continued YouTube dash-cam annotation, formalized with behavior tags and video evidence. The expanded framework organizes dangerous pedestrian patterns for annotation, simulation, and evaluation while aiming toward a more complete behavioral test space.
Problem
Behavior tags describe individual pedestrian actions but do not fully characterize recurring behavioral patterns relevant to autonomous-vehicle safety testing.
Method
The paper annotates dash-cam scenarios, reviews recurring behavior combinations against 12 existing archetypes, selects representative video evidence, and derives essential and optional behavior tags.
Results
The preprint introduces 7 additional pedestrian archetypes with formal definitions, behavior tags, comparisons, and representative video-frame evidence.
Takeaways & Limitations
The expanded archetype framework organizes connected dangerous pedestrian behaviors into models usable for annotation, simulation, and evaluation.
Abstract
from arXiv · showhide
In our prior work, Pedestrian Archetypes, we defined pedestrian archetypes as collections of behaviors that uniquely identify a specific type of pedestrian. The first paper proposed 12 pedestrian archetypes, including the Wanderer, Drunk, Distracted, Flash, Indecisive, Blind, Flock, Jaywalker, Elderly, Kid, Eventful, and Parked Pedestrian. These archetypes were introduced to move beyond single behavior labels and provide a more natural way to describe how dangerous pedestrians actually behave progressively in real-world traffic scenarios. However, upon further annotation of YouTube dash-cam videos, we identified 7 additional pedestrian archetypes with observable and significant behavioral differences from the previously proposed ones. These new archetypes capture pedestrian behavior patterns that could not be fully explained by the original taxonomy. In this pre-print, we introduce each new archetype, define its essential and optional behaviors, explain how it differs from previously proposed archetypes, and provide video-frame evidence showing the archetype in action.
I. INTRODUCTION & MOTIVATION
The paper argues that isolated behavior tags do not adequately represent recurring, dangerous pedestrian behavior patterns. It extends the prior taxonomy with seven additional archetypes identified through continued dash-cam annotation.
- Traditional AV safety models predict crossing, crossing location, and trajectory but do not fully capture complex, recurring patterns of dangerous pedestrian behavior.
- Behavior tags describe observable actions, whereas archetypes model broader pedestrian behavior patterns that may produce similar actions for different reasons.
- Archetypes group related essential and optional behaviors into coherent models for representing likely pedestrian actions in AV safety testing.
- Continued YouTube dash-cam annotation revealed recurring patterns that were visually observable and behaviorally distinct from the 12 original archetypes.
- The preprint introduces 7 additional archetypes with formal definitions, video evidence, related-archetype comparisons, and essential or optional behavior tags.
II. METHODOLOGY
The new archetypes were identified by systematically annotating risky pedestrian-vehicle interactions, reviewing recurring behavior combinations, and deriving essential and optional tags from the annotated dataset.
- The study identified new archetypes from YouTube dash-cam footage showing risky, near-miss, and collision pedestrian-vehicle interactions.
- The annotation process first applied the PedAnalyze ontology to observable actions including crossing, retreating, running, pausing, ignoring traffic, changing direction, and hesitating.
- Annotated scenarios were reviewed for recurring behavior combinations and compared with the 12 original archetypes to identify previously unmodeled patterns.
- Essential behaviors appeared in at least 40% of examples, optional behaviors in 10–39%, and behaviors below 10% were generally excluded unless qualitatively significant.
III. ARCHETYPES
The paper summarizes 12 previously defined pedestrian archetypes, each representing a recurring dangerous behavior pattern such as impairment, distraction, hesitation, or group movement.
- The Wanderer, Drunk, Distracted, Flash, Indecisive, Blind, Flock, Jaywalker, Elderly, Kid, Eventful, and Parked Pedestrian comprise the original taxonomy.
- The archetypes distinguish patterns including unpredictable lane wandering, impairment, reduced awareness, urgent sprinting, changing intent, and ignoring traffic.
- They also cover group movement, non-designated or red-light crossings, slow decisions, inexperience, involuntary external events, and interactions with parked vehicles.
A. The Con Artist
The Con Artist archetype describes staged pedestrian-vehicle collisions intended to obtain compensation or money, typically using slow or stationary vehicles to limit harm.
- A. The Con Artist: The Con Artist stages exaggerated collisions to claim insurance compensation or extort money from the driver.
- A. The Con Artist: The pedestrian waits in the vehicle’s path, jumps toward it as it slows, exaggerates the impact by pretending to fall, and receives assistance from an accomplice.
B. The Foreigner
The Foreigner is unfamiliar with local traffic norms and may misinterpret signals or intersection structures, leading to unsafe crossing decisions. A Venice example shows confusion at a 3-way intersection followed by an attempted crossing and retreat.
- The Foreigner misinterprets local traffic signals or structures because of unfamiliarity with local laws and norms.
- At a Venice 3-way intersection, a pedestrian attempts to cross, encounters an incoming vehicle, looks around, and retreats.
D. The Protester
The Protester occupies driving lanes while ignoring traffic regulations, disrupting traffic and potentially provoking aggressive driver responses. One cited example depicts a driver deliberately striking a protester.
- The Protester frequently ignores traffic regulations and occupies driving lanes.
- Occupying driving lanes can disrupt traffic and trigger aggressive responses from human drivers.
- A cited example shows a driver deliberately striking a protester with his truck.
F. The Pseudo Pedestrian
The Pseudo Pedestrian travels quickly and unpredictably on wheels, including rollerblades, skateboards, or wheelchairs. A skateboard example shows loss of control, a board entering traffic, and the pedestrian chasing it.
- The Pseudo Pedestrian moves fast and unpredictably via wheels, including rollerblading, skateboarding, or wheelchair use.
- A skateboarder loses balance after hitting the sidewalk, sending the skateboard into traffic and prompting a chase.
G. The Street Vendor
The Street Vendor approaches vehicles during slow or stopped traffic to market products. This behavior is risky because vendors remain close to moving cars and may approach the driver’s side with merchandise.
- The Street Vendor approaches cars during slow or stopped traffic to market products near vehicles.
- Street vendors walk alongside moving cars and approach the driver’s side with merchandise in hand.
IV. CONCLUSION
The preprint expands the pedestrian archetype taxonomy with seven new patterns that differ meaningfully from the original twelve. This broader taxonomy supports a more complete behavioral test space for autonomous vehicle safety and aims to prepare systems for rare, dangerous, and unpredictable pedestrians.
- IV. CONCLUSION: Seven new archetypes extend the framework by capturing observable pedestrian behavior patterns meaningfully different from the original twelve.The new archetypes were identified through further YouTube dash-cam video annotation.
- IV. CONCLUSION: Archetypes organize connected dangerous behaviors into communicable pedestrian models for annotation, simulation, and evaluation.
- IV. CONCLUSION: The expanded taxonomy moves toward a more complete behavioral test space for autonomous vehicles.
- IV. CONCLUSION: The long-term goal is to help AV systems prepare for rare, dangerous, and unpredictable pedestrians, not only law-abiding pedestrians.
APPENDIX
The appendix provides a link to the PedAnalyze pedestrian behavior tags documentation.
- APPENDIX: The appendix links to the PedAnalyze pedestrian behavior tags documentation.