MuSoHu: Multi-Modal Social Human Navigation (v2)

Last updated: 2026-10-14

Multi-Modal Social Human Navigation Dataset - A robotics research platform developed at George Mason University for collecting data on human-robot interactions in social navigation contexts.

Project Website: https://cs.gmu.edu/~xiao/Research/MuSoHu/

About

This project collects social human navigation data in natural human-inhabited public spaces. An egocentric data collection sensor suite worn by walking humans captures multi-modal robot perception data.

Dataset Scale

• ~100 km of navigation data
• 20 hours of recording
• 300 trials across various environments
• 13 human participants
• Multiple public spaces with natural social navigation interactions

Multi-Modal Sensor Suite

The wearable data collection device includes:

• 3D LiDAR - Robosense Helios 32 (360° environmental scanning)
• Stereo & Depth Camera - Stereolabs ZED 2i (visual perception and depth sensing)
• IMU - Inertial Measurement Unit (motion tracking)
• Odometry/Actions - Online processing for human navigation behavior extraction
• Microphone Array - Seeed Studio ReSpeaker Mic Array (spatial audio capture)

Documentation

For complete documentation, setup instructions, and production deployment guides, see:

• ROS 2 workspace - Sensor sources, SDK setup, build, bringup and evaluation recording
• Current work plan - Agreed ROS, odometry/IMU, recording and mechanical tasks
• Contributor instructions - Project context and working conventions
• Time synchronization - Offline phone/online NTP, software PTP and timestamp limits
• Depth/power benchmark - Recording power, neural depth modes and measurement scope
• Web recording - Recording presets, session metadata and dummy publishers
• Service deployment - Prepare and migrate the systemd service

Overview

MuSoHu is a platform for collecting data for social navigation research. It is built using ROS2 and includes a web interface for system interaction.

Features

• ROS2-based robotics platform
• Multi-sensor integration (ZED camera, LiDAR, IMU, Audio)
• Web interface for system control and monitoring
• Deployment with automatic restart capability
• Helmet-mounted sensor suite for data collection

Project Structure

• musohu_ws/ - ROS 2 Humble workspace with pinned sensor dependencies
  • src/helmet_bringup/ - Project launch files and sensor configurations
  • scripts/ - Dependency/SDK setup, symlink build, validation and recording helpers
• web-app/ - FastAPI/Uvicorn web interface for system control
• scripts/ - Setup, installation, and utility scripts
  • setup/ - Installation and configuration scripts
  • utils/ - Helper utilities and logging
  • udev_rules/ - Device access rules for sensors
  • hotspot/ - WiFi hotspot configuration
• docs/ - Complete documentation and guides