GLidE-SLAM

GL-accelerated Indirect-Direct Embedded SLAM

A hybrid visual SLAM system combining ORB-SLAM2's sparse backend with GPU-accelerated direct photometric tracking for embedded platforms.

Features

• Hybrid Tracking: Sparse indirect (ORB-SLAM2) backend with direct photometric tracking for intermediate frames
• GPU Acceleration: OpenGL ES 3.1 compute shaders for photometric tracking pipeline
• Embedded-First: Targets ARM platforms (Raspberry Pi 5, Radxa Zero 3W) with vendor-agnostic GPU support
• No Pangolin: Custom EGL/SDL-based viewer for embedded deployment
• Real-time: 3-4x speedup over CPU-only direct tracking on tested platforms

Installation

Quick Install (Debian 12 / Ubuntu 22.04+)

./install_dependencies.sh
./build.sh

Manual Installation

Prerequisites

Debian 12 (Bookworm) / Ubuntu 22.04+

sudo apt update
sudo apt install -y \
  build-essential cmake git \
  libeigen3-dev \
  libopencv-dev \
  libsdl2-dev \
  libboost-all-dev \
  libglm-dev \
  libegl1-mesa-dev \
  libgles2-mesa-dev

GPU Requirements

• OpenGL ES 3.1+ capable GPU
• For embedded platforms: Mali, VideoCore, or similar with compute shader support

Verify GPU support:

eglinfo | grep "OpenGL ES"
# Should show: OpenGL ES 3.1 or higher

Build

# Clone repository
git clone https://github.com/capsMD/GLidE-SLAM.git
cd GLidE-SLAM

# Build everything
./build.sh

This will:

1. Build DBoW2 in Thirdparty/DBoW2/
2. Build g2o in Thirdparty/g2o/
3. Extract vocabulary file
4. Build libGLidE_SLAM.so in lib/
5. Build example executables in Examples/Monocular/

Embedded Platforms (ARM)

Tested on:

• Radxa Zero 3W (RK3566, Mali-G52)
• Raspberry Pi 5 (BCM2712, VideoCore VII)

Additional notes:

• Mesa 24.x+ recommended for best Panfrost (Mali) support
• Kernel 6.1+ for optimal ARM GPU drivers

Usage

TUM Dataset Example

./Examples/Monocular/mono_tum \
  Vocabulary/ORBvoc.txt \
  Examples/Monocular/TUM1.yaml \
  /path/to/rgbd_dataset_freiburg1_xyz

KITTI Dataset Example

./Examples/Monocular/mono_kitti \
  Vocabulary/ORBvoc.txt \
  Examples/Monocular/KITTI00-02.yaml \
  /path/to/dataset/sequences/00

License

• Core SLAM framework (derived from ORB-SLAM2): GPL-3.0 (see License-gpl.txt)
• GLidE Engine (GPU acceleration): Proprietary (see source file headers)
  • Free for academic/research use with citation
  • Commercial licensing available

See LICENSE.txt and Dependencies.md for details.

Citation

If you use GLidE-SLAM in academic work, please cite:

[Your IROS 2026 paper citation - to be added]

Acknowledgments

This work builds upon ORB-SLAM2 by Raúl Mur-Artal and Juan D. Tardós.