koide3/glim

What it solves

GLIM provides a versatile and extensible framework for 3D mapping using range sensors. It addresses the need for high-accuracy mapping that remains consistent across different sensor types while offering the flexibility to manually correct mapping failures and integrate custom constraints.

How it works

The framework uses direct multi-scan registration error minimization on factor graphs to maintain consistency in mapping results. It supports GPU acceleration to increase processing speed and quality. To ensure versatility, it removes sensor-specific processes, allowing it to work with various range sensors such as spinning LiDAR, non-repetitive scan LiDAR, solid-state LiDAR, and RGB-D cameras. It also features a global callback slot mechanism that lets users access internal states and add additional constraints to the factor graph.

Who it’s for

It is designed for developers and researchers working with 3D mapping and localization, ranging from those using high-performance GPUs to those needing lightweight real-time mapping on low-specification hardware like a Raspberry Pi.

Highlights

• Sensor Agnostic: Compatible with various range sensors including spinning, non-repetitive, and solid-state LiDARs, as well as RGB-D cameras.
• Interactive Correction: Includes an interface for users to manually correct mapping failures and refine results.
• High Performance: Supports GPU acceleration for faster and more accurate scan matching.
• Extensible Architecture: Provides a callback mechanism for inserting custom constraints and accessing internal mapping states.