DuoRoad

A Dual-Participant Simulation Platform for
Driver–Worker Interaction and Multimodal Data Collection

Shuo Zhang · New York University (NYU)

PaperCode
Driver–worker interaction, multimodal human-behavior capture, vehicle sensing, and traffic co-simulation.
Driver–worker interaction, multimodal human-behavior capture, vehicle sensing, and traffic co-simulation.

Why?

Studying safety-critical road scenarios in the field is often infeasible or costly, yet these long-tail corner cases matter for human-factors and autonomous-driving research.

How?

DuoRoad brings two human participants—a driver and a VR worker—into one simulation for multimodal data collection and configurable scenarios.

Validation

A study with 33 participants evaluates the platform's usability.

Extensibility

Future extensions could study pedestrians and cyclists, human–AV interactions, and a wider range of safety-critical scenarios.

01

Interaction and Data Collection

Live VR workers and prerecorded human-motion animations coexist in a shared work zone.

Real-time VR workers and prerecorded human-motion animations.
02

Human Data Collection

Driver gaze and semantic segmentation, followed by VR motion and trajectory.

Driver data, then VR motion. VR footage and pose are separate recordings.
03

Multimodality Autonomous Driving Data Collection

Multi-view RGB cameras, LiDAR point clouds, vehicle trajectories, and driver control inputs.

04

Asynchronous Human-in-the-Loop Co-Simulation

CARLA–SUMO traffic simulation and independent data-collection streams.

05

Scenarios & Work-Zone Types

Six hazardous driving scenarios and three work-zone configurations.