dnn_attitude_estimation

System architecture

Installation

Build locally

If you are going to build it locally, install the requirements and follow the commands below.

Requirements

• ROS
• PyTorch

$ roscd
$ cd src
$ git clone https://github.com/ozakiryota/dnn_attitude_estimation
$ cd ..
$ catkin_make

Build with Docker

If you can use Docker, follow the commands below.

$ git clone https://github.com/ozakiryota/dnn_attitude_estimation
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./build.sh

Usage using 1 camera

Train DNN

Train the network with ozakiryota/image_to_gravity.

Run

$ roslaunch dnn_attitude_estimation camera_mle_inference.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./camera_mle_inference.sh

Open another terminal.

$ roslaunch dnn_attitude_estimation camera_mle_ekf_real.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./run.sh
$ roslaunch dnn_attitude_estimation camera_mle_ekf_real.launch

Citation

@ARTICLE{ozaki2021,
	author = {Ryota Ozaki and Yoji Kuroda},
	title = {EKF-based real-time self-attitude estimation with camera DNN learning landscape regularities},
	journal = {IEEE Robotics and Automation Letters (RA-L)},
	volume = {6},
	number = {2},
	pages = {1737--1744},
	year = {2021}
}

Usage using 4 cameras

Train DNN

Train the network with ozakiryota/multi_image_to_gravity.

Run

$ roslaunch dnn_attitude_estimation combined_cameras_mle_inference.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./combined_cameras_mle_inference.sh

Open another terminal.

$ roslaunch dnn_attitude_estimation combined_cameras_mle_ekf_real.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./run.sh
$ roslaunch dnn_attitude_estimation combined_cameras_mle_ekf_real.launch

Citation

Preparing ...

Usage using LiDAR

Train DNN

Train the network with ozakiryota/depth_image_to_gravity.

Run

$ roslaunch dnn_attitude_estimation lidar_regression_inference.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./lidar_regression_inference.sh

Open another terminal.

$ roslaunch dnn_attitude_estimation lidar_regression_ekf_real.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./run.sh
$ roslaunch dnn_attitude_estimation lidar_regression_ekf_real.launch

Citation

@ARTICLE{ozaki2021,
	author = {Ryota Ozaki and Naoya Sugiura and Yoji Kuroda},
	title = {LiDAR DNN based self-attitude estimation with learning landscape regularities},
	journal = {ROBOMECH Journal},
	volume = {8},
	number = {26},
	pages = {10.1186/s40648-021-00213-5},
	year = {2021}
}

Usage using 1 camera and LiDAR

Train DNN

Train the network with ozakiryota/color_and_depth_image_to_gravity.

Run

$ roslaunch dnn_attitude_estimation lidar_camera_regression_inference.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./lidar_camera_regression_inference.sh

Open another terminal.

$ roslaunch dnn_attitude_estimation lidar_camera_regression_ekf_real.launch
OR
$ cd dnn_attitude_estimation/docker/nvidia_docker1_kinetic
$ ./run.sh
$ roslaunch dnn_attitude_estimation lidar_camera_regression_ekf_real.launch

Citation

@Inproceedings{ozaki2021,
	author = {尾崎亮太 and 黒田洋司}, 
	title = {風景知識を学習するカメラ-LiDAR-DNNによる自己姿勢推定},
	booktitle = {第26回ロボティクスシンポジア予稿集},
	pages = {249--250},
	year = {2021}
}

Datasets

Some datasets are available at ozakiryota/dataset_image_to_gravity.

Trained models

Some trained models are available at dnn_attitude_estimation/keep.

Related repositories

• ozakiryota/gyrodometory
• ozakiryota/msg_conversion
• ozakiryota/msg_printer
• ozakiryota/velodyne_pointcloud_to_depthimage