VP-SOM

Authors: [Zhang Jiadong], [Wang Wei]

VP-SOM is a novel View Planning Method for Indoor Sparse Object Model based on Information Abundance and Observation Continuity.

1. Prerequisites

We have tested the system in Ubuntu 18.04. + ROS melodic for Motion module and the connection between Motion module&View-Planning module. We suggest installing the "desktop-full" version of ROS. + Prerequisites of Active SLAM are the same as ORB_SLAM2, including C++11, OpenCV, Eigen3.4.10, Eigen3, Pangolin, DBoW2, g2o and PCL1.8. + Gazebo 9.0 for Simulization environment. + Rviz for visulization of object map, camera view and robot motion.

2. Building

• Create ROS workspace and download our package:

  cd {ROS_WORKSPACE}/src
  git clone https://github.com/TINY-KE/VP-SOM.git
  cd VP-SOM
  

• Complie the thirdparty libraries of Active SLAM:

  cd Active_SLAM_based_on_VP-SOM
  chmod +x build_thirdparty.sh       
  ./build_thirdparty.sh
  

• Complie the Active SLAM and YOLO:

  cd {ROS_WORKSPACE}
  catkin_make
  

3. Usage of view planning method

There are many parameters in the file "config/kinectv1.yaml" that can affect VP-SOM. This section will introduce these parameters.

• [PubGlobalGoal] :
• [PubLocalGoal] :
• [MAM.Reward_angle_cost] :
• [MAM.Reward_dis] :
• [Planle.Safe_radius] :
• [ConstraintType] :
• [ObserveMaxNumBackgroudObject] :
• [IE.ThresholdEndMapping] :
• [Plane.Height.Max] and [Plane.Height.Min] :
• [IE.PublishIEwheel] :
• [IE.P_occ], [IE.P_free], [IE.P_prior] :
• [IE.ThresholdPointNum] :
• Series of [Trobot_camera] :
• Series of [Tworld_camer] :
• Other parameters have little effect and will be updated in the future.

4. Simulization Environment and Fabo robot

• 1. Simulization Environment

  • One backgroud object

    roslaunch ASLAM_gazebo_world hokuyo_kinectv1_bringup_moveit_1bo.launch
    

  • Two backgroud object

    roslaunch ASLAM_gazebo_world hokuyo_kinectv1_bringup_moveit_2bo.launch 
    

  • Three backgroud object

    roslaunch ASLAM_gazebo_world hokuyo_kinectv1_bringup_moveit_3bo.launch 
    

    

• 2. Fabo robot controller

  • Control robot by keyboard. This corresponds to manual mode where PubGlobalGoal=0. Press "IJLK," to control the movement of the chassis. Press "G" to publish a signal that has arrived at the NBV to the view-planning program to start a new round of view-planning

    roslaunch fabo_teleop fabo_teleop.launch
    

  • Robot move autonomously by MoveIt and 2D grid map. This corresponds to autonomous mode where PubGlobalGoal=1.

    roslaunch fabo_robot_gazebo  fake_navigation.launch
    

5. Test the Active SLAM

• 1. Start simulation environment and robot controller as the section 4
• 2. YOLO object detection

  roslaunch darknet_ros darknet_kinectv1.launch
  

• 3. Active SLAM

  roslaunch active_slam_vpsom aslam.launch
  

  • Visulization of object map, camera view and robot motion roslaunch active_slam_vpsom rviz.launch

6. Evaluation

The results of sparse object map and observation trajectories of different view-planning methods are saved in "eval/temp". Evaluate various methods by comparing sparse object maps and observation trajectories.

rosrun active_slam_vpsom eval 

The groudtruth of object models in the simulation environments can be extracted from the "world" file of gazebo

rosrun active_eao_new extract_gazebo_groudth  [the path of gazebo world file]

Other details to be updated later.