Noisy Global Map using ZEDm causing obstacle avoidance navigation problem

Posted by seanxu112 on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Noisy-Global-Map-using-ZEDm-causing-obstacle-avoidance-navigation-problem-tp6039.html

Hi,

Sorry to have another problem so soon. Localization with laser and ZEDm camera works great right now. However, it seems like the global map has a lot of salt and pepper noise. I tried to solve it by post processing it with rtabmap-databaseViewer, but it seems like we cannot save the new processed data base after refining links and detect loop closure. I also applied cloud filtering with 0.1meter of 100 points. There is just one single point on top of the map, and move_base is registering it as an obstacle and trying to avoid it. Also, my cloud map topic and cloud obstacles topic seems to be publishing the same thing.

I thought about using mark_threshold in costmap params yaml file. Here is my local and global costmap yaml:

local_costmap:
  global_frame: /map
  robot_base_frame: /base_link
  update_frequency: 2.0
  publish_frequency: 2.0
  static_map: false
  rolling_window: true
  width: 6.0
  height: 6.0
  resolution: 0.05

global_costmap:
  global_frame: /map
  robot_base_frame: /base_link
  update_frequency: 2.0
  static_map: true

costmap_common_params:
obstacle_range: 2.5
raytrace_range: 3.0
max_obstacle_height: 1.0
footprint: [[0.36, 0.29], [0.36, -0.29], [-0.31, -0.29], [-0.31, 0.29]]
#robot_radius: ir_of_robot
inflation_radius: 0.05

mark_threshold: 10

transform_tolerance: 0.5
observation_sources: laser_scan_sensor point_cloud_sensor #laser_scan_sensor point_cloud_sensor

laser_scan_sensor: {sensor_frame: laser_frame, data_type: LaserScan, topic: /scan, marking: true, clearing: true}

point_cloud_sensor: {sensor_frame: zed_camera_center, data_type: PointCloud2, topic: /rtabmap/cloud_map, marking: true, clearing: true}


Is there any way to solve this with move_base or post-processing the database?

Sincerely,

Sean