Ultra-fast elevation mapping on embedded robots — 100+ Hz on Jetson Orin

<a href="#start-with-ros1">ROS1</a> · <a href="#start-with-ros2">ROS2</a> · <a href="#use-fastdem-as-a-c-library">C++ Library</a> <img src="assets/fastdem_laser_local.gif" width="49%" alt="FastDEM real-time elevation mapping with LiDAR point cloud" /> <img src="assets/fastdem_rgbd_local.gif" width="48.3%" alt="FastDEM elevation mapping with RGB-D camera" /> <img src="assets/fastdem_rgbd_global.gif" width="97.5%" alt="FastDEM global elevation mapping in outdoor environment" />

FastDEM is a lightweight C++17 library that builds dense elevation maps from LiDAR and RGB-D point clouds in real time, without GPU. It supports both robot-centric local mapping and fixed-origin global mapping, with built-in post-processing for terrain analysis.

See also:

LeSTA — Self-supervised traversability learning for mobile robots (RA-L 2024)

Features

Fast — 100+ Hz on Jetson Orin. ~10ms per scan, on CPU alone.
Lightweight — Just Eigen at core. No PCL, OpenCV, or CUDA required.
ROS-agnostic — Clean C++ API, with optional ROS support.
Sensor-Aware — Physics-based sensor models for LiDAR and RGB-D range measurements.
Multiple Estimators — Kalman Filter (parametric), P² Quantile estimator (non-parametric).
Local + Global Mapping — Robot-centric or map-centric terrain mapping.
Post-processing — Raycasting, Uncertainty fusion, Inpainting, Feature extraction, and more.

Performance

The mapping pipeline runs at ~10 ms on embedded CPUs — fast enough to leave ample room for post-processing.

Measured with Velodyne VLP-16 (~30K pts/scan) · 15×15 m map at 0.1 m resolution

Dependencies

Eigen3, yaml-cpp, spdlog
nanoGrid, nanoPCL — bundled automatically

Start with ROS1

Prerequisites: Ubuntu 20.04, ROS Noetic

# Dependencies
sudo apt install libeigen3-dev libyaml-cpp-dev libspdlog-dev
sudo apt install ros-noetic-tf2-eigen ros-noetic-grid-map-msgs

# Clone and build
cd ~/catkin_ws/src
git clone https://github.com/Ikhyeon-Cho/FastDEM.git
catkin build fastdem_ros

# Run (add global_mapping:=true for map-centric mode)
roslaunch fastdem_ros run.launch rviz:=true

Configuration: ros1/config/local_mapping.yaml · ros1/config/global_mapping.yaml

Start with ROS2

Prerequisites: Ubuntu 22.04, ROS2 Humble

Other ROS2 distributions may also work but are not yet tested.

# Dependencies
sudo apt install libeigen3-dev libyaml-cpp-dev libspdlog-dev
sudo apt install ros-humble-tf2-eigen ros-humble-grid-map-msgs

# Clone and build
cd ~/ros2_ws/src
git clone https://github.com/Ikhyeon-Cho/FastDEM.git
colcon build --packages-up-to fastdem_ros2

# Run (add global_mapping:=true for map-centric mode)
ros2 launch fastdem_ros2 run.launch.py rviz:=true

Configuration: ros2/config/local_mapping.yaml · ros2/config/global_mapping.yaml

Use FastDEM as a C++ Library

FastDEM can be used without ROS as a standalone C++ library.

# Dependencies
sudo apt install libeigen3-dev libyaml-cpp-dev libspdlog-dev

# Clone and build
git clone https://github.com/Ikhyeon-Cho/FastDEM.git
cd FastDEM/fastdem
mkdir build && cd build
cmake ..
make -j$(nproc)
sudo make install

#include <fastdem/fastdem.hpp>

fastdem::ElevationMap map;
map.setGeometry(15.0, 15.0, 0.1);  // width, height, resolution [m]

auto cfg = fastdem::loadConfig("config/default.yaml");
fastdem::FastDEM mapper(map, cfg);

// With explicit transforms
mapper.integrate(cloud, T_base_sensor, T_world_base);

Citation

FastDEM was originally developed for the following research:

'Learning Self-supervised Traversability with Navigation Experiences of Mobile Robots' IEEE Robotics and Automation Letters (RA-L), 2024

@article{cho2024learning,
  title={Learning Self-Supervised Traversability With Navigation Experiences of Mobile Robots: A Risk-Aware Self-Training Approach},
  author={Cho, Ikhyeon and Chung, Woojin},
  journal={IEEE Robotics and Automation Letters},
  year={2024},
  volume={9},
  number={5},
  pages={4122-4129},
  doi={10.1109/LRA.2024.3376148}
}

Contact: ikhyeon.c@gmail.com

</div>

FastDEM

Install / Use

README