CLID-SLAM: A Coupled LiDAR-Inertial Neural Implicit Dense SLAM With Region-Specific SDF Estimation

TL;DR: CLID-SLAM is a tightly-coupled LiDAR-Inertial Odometry and dense Mapping framework, which utilizes the Iterated Error-State Extended Kalman Filter (IESEKF) to fuse Signed Distance Function (SDF) predictions and IMU data to improve robustness.

Installation

Platform Requirements

We tested our code on Ubuntu 20.04 with an NVIDIA RTX 4090.

Steps

# Install Miniforge3
wget -O Miniforge3.sh "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh"
bash Miniforge3-$(uname)-$(uname -m).sh
# Clone the repo
git clone [email protected]:DUTRobot/CLID-SLAM.git
cd CLID-SLAM
# create conda environment
mamba create -n ros_env python=3.11
# Install ROS noetic
mamba install ros-noetic-desktop-full -c robostack-noetic -c conda-forge
# install pytorch
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
pip3 install -r requirements.txt

Data Preparation

Download ROSbag Files

Newer College Dataset (also available on Baidu Disk due to Google Drive's maximum file size limit).

SubT-MRS Dataset

Convert to Sequences

Edit ./dataset/converter/config/rosbag2dataset.yaml.

input_bag: '/home/jiang/hku1_converted.bag'
output_folder: './dataset/hku/'
imu_topic: '/livox/imu'
lidar_topic: '/livox/pointcloud2'
image_topic: "/left_camera/image/compressed"  # or /camera/color/image_raw
batch_size: 100 # Number of messages per batch
end_frame: -1 # -1 means process the entire bag file

Run:

python3 ./dataset/converter/rosbag2dataset_parallel.py

How to run it

python3 slam.py ./config/run_ncd128.yaml

Acknowledgements

This project builds on PIN-SLAM by PRBonn/YuePanEdward. We gratefully acknowledge their valuable contributions.