Direct LiDAR Odometry: Fast Localization with Dense Point Clouds

Related tags

Deep Learningdirect_lidar_odometry
Overview

Direct LiDAR Odometry: Fast Localization with Dense Point Clouds

DLO is a lightweight and computationally-efficient frontend LiDAR odometry solution with consistent and accurate localization. It features several algorithmic innovations that increase speed, accuracy, and robustness of pose estimation in perceptually-challenging environments and has been extensively tested on aerial and legged robots.

This work was part of NASA JPL Team CoSTAR's research and development efforts for the DARPA Subterranean Challenge, in which DLO was the primary state estimation component for our fleet of autonomous aerial vehicles.


drawing drawing

drawing

Instructions

DLO requires an input point cloud of type sensor_msgs::PointCloud2 with an optional IMU input of type sensor_msgs::Imu. Note that although IMU data is not required, it can be used for initial gravity alignment and will help with point cloud registration.

Dependencies

Our system has been tested extensively on both Ubuntu 18.04 Bionic with ROS Melodic and Ubuntu 20.04 Focal with ROS Noetic, although other versions may work. The following configuration with required dependencies has been verified to be compatible:

  • Ubuntu 18.04 or 20.04
  • ROS Melodic or Noetic (roscpp, std_msgs, sensor_msgs, geometry_msgs, pcl_ros)
  • C++ 14
  • CMake >= 3.16.3
  • OpenMP >= 4.5
  • Point Cloud Library >= 1.10.0
  • Eigen >= 3.3.7

Installing the binaries from Aptitude should work though:

sudo apt install libomp-dev libpcl-dev libeigen3-dev

Compiling

Create a catkin workspace, clone the direct_lidar_odometry repository into the src folder, and compile via the catkin_tools package (or catkin_make if preferred):

mkdir ws && cd ws && mkdir src && catkin init && cd src
git clone https://github.com/vectr-ucla/direct_lidar_odometry.git
catkin build

Execution

After sourcing the workspace, launch the DLO odometry and mapping ROS nodes via:

roslaunch direct_lidar_odometry dlo.launch \
  pointcloud_topic:=/robot/velodyne_points \
  imu_topic:=/robot/vn100/imu

Make sure to edit the pointcloud_topic and imu_topic input arguments with your specific topics. If IMU is not being used, set the dlo/imu ROS param to false in cfg/dlo.yaml. However, if IMU data is available, please allow DLO to calibrate and gravity align for three seconds before moving. Note that the current implementation assumes that LiDAR and IMU coordinate frames coincide, so please make sure that the sensors are physically mounted near each other.

Test Data

For your convenience, we provide example test data here (9 minutes, ~4.2GB). To run, first launch DLO (with default point cloud and IMU topics) via:

roslaunch direct_lidar_odometry dlo.launch

In a separate terminal session, play back the downloaded bag:

rosbag play dlo_test.bag

drawing

Citation

If you found this work useful, please cite our manuscript:

@article{chen2021direct,
  title={Direct LiDAR Odometry: Fast Localization with Dense Point Clouds},
  author={Chen, Kenny and Lopez, Brett T and Agha-mohammadi, Ali-akbar and Mehta, Ankur},
  journal={arXiv preprint arXiv:2110.00605},
  year={2021}
}

Acknowledgements

We thank the authors of the FastGICP and NanoFLANN open-source packages:

  • Kenji Koide, Masashi Yokozuka, Shuji Oishi, and Atsuhiko Banno, “Voxelized GICP for Fast and Accurate 3D Point Cloud Registration,” in IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2021, pp. 11 054–11 059.
  • Jose Luis Blanco and Pranjal Kumar Rai, “NanoFLANN: a C++ Header-Only Fork of FLANN, A Library for Nearest Neighbor (NN) with KD-Trees,” https://github.com/jlblancoc/nanoflann, 2014.

License

This work is licensed under the terms of the MIT license.


drawing drawing

drawing drawing

Comments
  • Coordinate system conversion

    Coordinate system conversion

    Hi, thanks for your great work. I used my lidar and imu to record the data set and run it in DLO. The following figure is the map generated by the algorithm. I found that IMU drift is very serious. Is this related to coordinate system transformation? How can I set up and configure files to improve this situation? d0724d53975aa58f749b1abfa692456 3ae5711ad4dc74f03c683af98bc7db9

    opened by HomieRegina 15
  • HI, I use the nanoicp to location, but it cann`t work

    HI, I use the nanoicp to location, but it cann`t work

    Thanks for your work, I want to use nanoicp to location, it cann`t work right, but I use the pcl Gicp， it can work, so, there is anything i need to do?

    opened by tust13018211 10
  • Map generation

    Map generation

    Hi, How is the map generated? What algorithm is used? Can we upload a pcd map of the environment and then apply dlo just for localisation? I assumed dlo as a localisation algorithm, but it seems to take in the point cloud data from the bag file and generate a map on its own.

    opened by Srichitra-S 9
  • gicp speed

    gicp speed

    I run source code each scan's gicp process just cost 2-3 ms, when i run my own code use gicp library it cost 100-200ms. Both Input cloud's size almost same, which setting options may cause this phenomenon?

    opened by yst1 8
  • how to localize on a given pcd map?

    how to localize on a given pcd map?

    i have a trouble understanding how to localise on a given 3d pcd map, can anyone explain the steps

    i have a pcd map which i can load into rviz, and i have lidar-points and imu data, then how to use this package for localization?

    opened by srinivasrama 6
  • jacobian

    jacobian

    Hello author, I want to derive the jacobian in the code. I see we use global perturbation to update. When i try this, i meet this problem to look for your help. image When using Woodbury matrix identity it seems to make the formula more complicated, so i'm stuck to look for your help. Thanks for your help in advance!

    Best regards Xiaoliang jiao

    opened by narutojxl 6
  • the transformation between your lidar and imu?

    the transformation between your lidar and imu?

    I use a vlp16 lidar and a microstran imu(3DM-GX5-25) to run your program. But I failed. I think my transformation between imu and lidar is different from yours. So could you tell your true transformation between imu and lidar. the following pitcure is my configuration: 453670779 Thank you very much!

    opened by nonlinear1 6
  • Conversion of lidar and IMU coordinate system

    Conversion of lidar and IMU coordinate system

    Hi! I use a vlp16 lidar and a microstran imu(3DM-GX5-25) to record bags. And here is the part of the map I built with your algorithm. Because the coordinate systems of IMU and lidar are not consistent in the physical direction, this has affected the construction of the two-layer environment. Could you please tell me how to convert IMU and lidar coordinate system? 图片

    opened by HomieRegina 4
  • some trivial questions

    some trivial questions

    Hello authors, I have some trivial questions to look for your help. Thanks for your help.

    opened by narutojxl 4
  • Coordinate Frame

    Coordinate Frame

    Hi,

    Thank you for sharing this great work.

    I've set of problem such as defining the coordinate system of the odometry . Which coordinate frame do you use for odometry NED or BODY. I thought you use NED coordinate frame so that I expected when I move Lidar + IMU to the North direction, x should be positive increased but it has not been stable behaviour. I also moved the system to the East direction it should be caused positive increase on y axis but same problem. I think you use BODY frame or something different coordinate frame for odometry.

    I wonder also how do you send odometry information to the autopilot of drone. I prefered mavros and selected the related odometry ros topic and send them without change so I saw the odometry message in the autopilot but when odometry receive to the autopilot of drone it looks like x and y exchanged in autopilot even they are not exchanged odometry output of direct lidar odometry.

    Summary of the Question is that which coordinate frame do you use for odometry ? How can I set up the odometry frame to NED? x and y are negative decreased when I move lidar+imu to the North and the East respectively. and Z looks like 180 reversed. Should I multiply x and y with minus or apply rotation matrix to fix negative decreasing x, y odometry ?

    How can I ensure that autopilot and your odometry coordinate plane are in the same coordinate plane ?

    opened by danieldive 3
  • How to correct point cloud caused by motion？

    How to correct point cloud caused by motion？

    Hi, thank you for your great work about DLO. I have a question. When the robot is moving fast, the lidar point cloud will generate distortion due to movement, which will cause negative impact on the construction of the map. So how did DLO deal with this problem?

    opened by JACKLiuDay 2
Releases(v1.4.2)
Owner
VECTR at UCLA
Verifiable & Control-Theoretic Robotics Laboratory
VECTR at UCLA
GitHub Repository
Sign Language Translation with Transformers (COLING'2020, ECCV'20 SLRTP Workshop)

transformer-slt This repository gathers data and code supporting the experiments in the paper Better Sign Language Translation with STMC-Transformer.

Kayo Yin 107 Dec 27, 2022
Library for converting from RGB / GrayScale image to base64 and back.

Library for converting RGB / Grayscale numpy images from to base64 and back. Installation pip install -U image_to_base_64 Conversion RGB to base 64 b

Vladimir Iglovikov 16 Aug 28, 2022
DiffSinger: Singing Voice Synthesis via Shallow Diffusion Mechanism (SVS & TTS); AAAI 2022; Official code

DiffSinger: Singing Voice Synthesis via Shallow Diffusion Mechanism This repository is the official PyTorch implementation of our AAAI-2022 paper, in

Jinglin Liu 803 Dec 28, 2022
Python Interview Questions

Python Interview Questions Clone the code to your computer. You need to understand the code in main.py and modify the content in if __name__ =='__main

ClassmateLin 575 Dec 28, 2022
[NeurIPS 2021] Code for Unsupervised Learning of Compositional Energy Concepts

Unsupervised Learning of Compositional Energy Concepts This is the pytorch code for the paper Unsupervised Learning of Compositional Energy Concepts.

45 Nov 30, 2022
Local Similarity Pattern and Cost Self-Reassembling for Deep Stereo Matching Networks

Local Similarity Pattern and Cost Self-Reassembling for Deep Stereo Matching Networks Contributions A novel pairwise feature LSP to extract structural

31 Dec 06, 2022
Heat transfer problemas solved using python

heat-transfer Heat transfer problems solved using python isolation-convection.py compares the temperature distribution on the problem as shown in the

2 Nov 14, 2021
Real-time object detection on Android using the YOLO network with TensorFlow

TensorFlow YOLO object detection on Android Source project android-yolo is the first implementation of YOLO for TensorFlow on an Android device. It is

Nataniel Ruiz 624 Jan 03, 2023
Graph Neural Networks with Keras and Tensorflow 2.

Welcome to Spektral Spektral is a Python library for graph deep learning, based on the Keras API and TensorFlow 2. The main goal of this project is to

Daniele Grattarola 2.2k Jan 08, 2023
Proto-RL: Reinforcement Learning with Prototypical Representations

Proto-RL: Reinforcement Learning with Prototypical Representations This is a PyTorch implementation of Proto-RL from Reinforcement Learning with Proto

Denis Yarats 74 Dec 06, 2022
[CVPR 2022] CoTTA Code for our CVPR 2022 paper Continual Test-Time Domain Adaptation

CoTTA Code for our CVPR 2022 paper Continual Test-Time Domain Adaptation Prerequisite Please create and activate the following conda envrionment. To r

Qin Wang 87 Jan 08, 2023
Digital Twin Mobility Profiling: A Spatio-Temporal Graph Learning Approach

Digital Twin Mobility Profiling: A Spatio-Temporal Graph Learning Approach This is the implementation of traffic prediction code in DTMP based on PyTo

chenxin 1 Dec 19, 2021
Justmagic - Use a function as a method with this mystic script, like in Nim

justmagic Use a function as a method with this mystic script, like in Nim. Just

witer33 8 Oct 08, 2022
Adversarial Robustness Comparison of Vision Transformer and MLP-Mixer to CNNs

Adversarial Robustness Comparison of Vision Transformer and MLP-Mixer to CNNs ArXiv Abstract Convolutional Neural Networks (CNNs) have become the de f

Philipp Benz 12 Oct 24, 2022
Do you like Quick, Draw? Well what if you could train/predict doodles drawn inside Streamlit? Also draws lines, circles and boxes over background images for annotation.

Streamlit - Drawable Canvas Streamlit component which provides a sketching canvas using Fabric.js. Features Draw freely, lines, circles, boxes and pol

Fanilo Andrianasolo 325 Dec 28, 2022
Framework to build and train RL algorithms

RayLink RayLink is a RL framework used to build and train RL algorithms. RayLink was used to build a RL framework, and tested in a large-scale multi-a

Bytedance Inc. 32 Oct 07, 2022
PyTorch Implementation of Unsupervised Depth Completion with Calibrated Backprojection Layers (ORAL, ICCV 2021)

Unsupervised Depth Completion with Calibrated Backprojection Layers PyTorch implementation of Unsupervised Depth Completion with Calibrated Backprojec

80 Dec 13, 2022
Implementation of the ivis algorithm as described in the paper Structure-preserving visualisation of high dimensional single-cell datasets.

Implementation of the ivis algorithm as described in the paper Structure-preserving visualisation of high dimensional single-cell datasets.

beringresearch 285 Jan 04, 2023
Codebase for arXiv preprint "NeRF++: Analyzing and Improving Neural Radiance Fields"

NeRF++ Codebase for arXiv preprint "NeRF++: Analyzing and Improving Neural Radiance Fields" Work with 360 capture of large-scale unbounded scenes. Sup

Kai Zhang 722 Dec 28, 2022
Hidden-Fold Networks (HFN): Random Recurrent Residuals Using Sparse Supermasks

Hidden-Fold Networks (HFN): Random Recurrent Residuals Using Sparse Supermasks by Ángel López García-Arias, Masanori Hashimoto, Masato Motomura, and J

Ángel López García-Arias 4 May 19, 2022