3D Dynamic Mapping For Autonomous Robots:

A real-time dynamic obstacle tracking and mapping system for UAV navigation and collision avoidance with an RGB-D camera

This repo contains 3D occupancy map, dynamic obstacle detection and osbatcle trajectory prediction for autonomous navigation. The dynamic obstacles will be detected, tracked and merged into the static occupancy map with a corresponding trajectory prediction (ICRA2023)

paper link: A real-time dynamic obstacle tracking and mapping system for UAV navigation and collision avoidance with an RGB-D camera

Youtube Demo Video Link: https://www.youtube.com/watch?v=u5zblVx8KRc

https://user-images.githubusercontent.com/60746163/199142937-e9493b8b-fb23-4d75-b5fc-f5c48da14e91.mp4

You may want to link Eigen to defalt location if Eigen is not found

sudo ln -s /usr/include/eigen3/Eigen/ /usr/include/Eigen

I. Install

prerequisite: Eigen, OpenCV
git clone https://github.com/Shawn207/map_manager_pub.git
cd ~/catkin_ws
catkin_make

II. Run demo

Please specify the camera parameters and the depth image/pose topic in map_manager_pub/map_manager/cfg/dynamic_map_param, then run the dynamic map launch:

roslaunch map_manager dynamic_map.launch
roslaunch map_manager rviz.launch

III. Parameters

Please find parameters in map_manager_pub/map_manager/cfg/dynamic_map_param.yaml files. The parameters are explained by comments.

IV. ROS Topics

Subsribe the following topics for occupancy and ESDF map:

• Localization topic: robot/odometry or robot/pose (please enter the name of your topic in the parameter files)
• Depth camera topic: camera/depth (defined in the config file)

Publish the following topics:

• occupancy map visualization: dynamic_map/inflated_voxel_map
• dynamic obstacles bounding boxes: dynamic_map/box_visualization_marker
• obstacle trajectory prediciton: dynamic_map/traj_marker

V. C++ Code API

Example code:

#include <map_manager/dynamicMap.h>

int main(){
  ...
  map_manager::dynamicMap m;
  m.initMap(nh);

  // collision checking with static obstacles given a point
  Eigen::Vector3d pos (1.0, 1.0, 1.0)
  bool hasCollision = m.isOccupied(pos);

  // get dynamic obstacles information(poseition, velocity, size)
  std::vector<Eigen::Vector3d> obstaclesPos;
  std::vector<Eigen::Vector3d> obstaclesVel;
  std::vector<Eigen::Vector3d> obstaclesSize;
  m.getDynamicObstacles(obstaclesPos, obstaclesVel, obstaclesSize);

  // get predicted trajectory
  std::vector<std::vector<std::vector<geometry_msgs::Point>>> predTraj;
  m.getTPredTraj(predTraj);

  ...
}

VI. Reference

If you find this work useful, please cite the paper:

@article{xu2022real,
  title={A real-time dynamic obstacle tracking and mapping system for UAV navigation and collision avoidance with an RGB-D camera},
  author={Xu, Zhefan and Zhan, Xiaoyang and Chen, Baihan and Xiu, Yumeng and Yang, Chenhao and Shimada, Kenji},
  journal={arXiv preprint arXiv:2209.08258},
  year={2022}
}