Home - uiuc-iml/IML-Perception-Box GitHub Wiki

Welcome to the IML-Perception-Box wiki!

Perception Box Overview

The Perception Box is an integrated SLAM and 3D semantic mapping system designed for real-time spatial understanding. It combines visual-inertial SLAM with semantic and metric reconstruction to generate labeled 3D maps of the environment. The system supports live operation over XML-RPC, enabling control and data retrieval from networked clients.

Components

• SLAM Module
  Performs real-time visual-inertial odometry, pose tracking, and loop closure.

• Mapping Module
  Builds a 3D map from SLAM output, with optional semantic labeling and color integration.

System Coordination

These modules operate in coordination, with the SLAM system providing pose and frame data to the mapping module. The Perception Box supports starting, pausing, resuming, and stopping the mapping process via API calls. It also allows retrieval of semantic and metric maps in both point cloud and mesh formats.

Table of Contents

• Acknowledgments
• Features

Acknowledgments

We extend our gratitude to our mentors and advisors for their guidance and support:

• Professor Kris Hauser
• João Marques

This project is built as a fork of the stella-vslam-examples repository, with contributions from members of the IML Research Team:

• Aaditya Voruganti (GitHub)
• Vallabh Nadgir (GitHub)

Features

• Multi-Camera Support: Monocular, and RGB-D cameras.
• Flexible Input: Supports video files and live camera feeds.
• Pangolin Viewer: OpenGL-based viewer for real-time visualization.
• Optimized for Embedded Systems: Suitable for NVIDIA Jetson platforms.

---