Dynamic-VINS

RGB-D Inertial Odometry for a Resource-restricted Robot in Dynamic Environments

1. Introduction

Dynamic-VINS is a real-time RGB-D Inertial Odometry system for resource-restricted robots in dynamic environments.

• Dynamic feature recognition by object detection and depth information with the performance comparable to semantic segmentation.
• Grid-based feature detection and efficient high-quality FAST feature extraction.
• Competitive localization accuracy and robustness in dynamic environments are shown in a real-time application on resource-restricted platforms, such as HUAWEI Atlas200 DK, NVIDIA Jetson AGX Xavier.

Authors: Jianheng Liu, Xuanfu Li, Yueqian Liu, and Haoyao Chen from the Networked RObotics and Sytems Lab, HITSZ

If you use Dynamic-VINS for your academic research, please cite the following paper [pdf].

@ARTICLE{9830851,  
  author={Liu, Jianheng and Li, Xuanfu and Liu, Yueqian and Chen, Haoyao},  
  journal={IEEE Robotics and Automation Letters},  
  title={RGB-D Inertial Odometry for a Resource-Restricted Robot in Dynamic Environments},   
  year={2022},  
  volume={7},  
  number={4},  
  pages={9573-9580},  
  doi={10.1109/LRA.2022.3191193}}

1.1. Framework

1.2. Related Video:

Video links: Youtube or Bilibili.

2. Installation

Tested on Ubuntu 18.04 and 20.04.

Find how to install Dynamic-VINS and its dependencies here: Installation instructions.

3. Run datasets examples

3.1. OpenLORIS

Download OpenLORIS datasets.

Take OpenLORIS-cafe as examples.

tar -xzvf cafe1-1_2-rosbag.tar
cd cafe
rosbag decompress cafe*
python YOUR_PATH_TO_DYNAMIC_VINS/scripts/merge_imu_topics.py cafe1-1.bag cafe1-2.bag

NVIDIA devices (pytorch)

roslaunch vins_estimator openloris_vio_pytorch.launch
roslaunch vins_estimator vins_rviz.launch # Visualization
rosbag play YOUR_PATH_TO_DATASET/cafe.bag 

NVIDIA devices (tensorrt)

roslaunch vins_estimator openloris_vio_tensorrt.launch
roslaunch vins_estimator vins_rviz.launch # Visualization
rosbag play YOUR_PATH_TO_DATASET/cafe.bag 

HUAWEI Atlas200DK

roslaunch vins_estimator openloris_vio_atlas.launch

Running Dynamic-VINS on HUAWEI Atlas200DK requires multile devices communication setting. For specific instructions, please refer to the MULTIPLE_DEVICES. And other kinds of edge devices also could refer to this instruction.

3.2. HITSZ & THUSZ Datasets

Please prepare enough space for the datasets.

• HITSZ(41.0GB x 2)
• THUSZ(51.3GB x 2)

1. download datasets vis Dyanmic-VINS-Datasets.
2. run following cmd:

# bash scripts/download_hitsz.sh # bash scripts/download_thusz.sh
python3 scipts/rosbag_merge_chunk.py Datasets/hitsz_00.bag # python3 scipts/rosbag_merge_chunk.py Datasets/hitsz_00.bag
# rm ./Datasets/hitsz_*.bag # rm ./Datasets/thusz_*.bag
roslaunch vins_estimator realsense_vio_campus.launch
roslaunch vins_estimator vins_rviz.launch
rosbag play Datasets/hitsz.bag # rosbag play thusz.bag 

4. Configuration

• Detailed illustration of configuration please refer to realsense configuration.
• A running procedure across two different platform is exampled in Multiple Devices running_procedure.
• HUAWEI Atlas200DK setup please refer to HUAWEI Atlas200DK Setup.

5. Acknowledgments

Dynamic-VINS is extended based on VINS-Mono, VINS-RGBD, yolov5, tensorrt_yolov5, ascend_yolo.