Revision history [back]

The launch file you posted will do two things:

• Load the saved map and publish it on the /map topic.
• Run the costmap_2d_node, which will output an occupancy grid for other nodes to use.

However, there are still quite a few other things that need to be set up to do what you want.

• Run amcl to localize on the map.
• Generate an odometry estimate
• Configure layers for the costmap, which will populate the occupancy grid with the data from the saved map.

First, localization. The costmap node is looking for a transform from base_footprint frame to map frame, not a base_footprint topic. This transform represents the robot's best guess of where it is on the map. The process of computing a position guess is known as localization. The most widely-used and mature ROS package to do map based localization is amcl. It uses the raw LIDAR scan, and finds the best match to the loaded map to compute the most probable position of the robot.

However, amcl needs an odometry estimate, which needs to be generated by some other node in the system. This odometry node would integrate measured speed and yaw rate to generate a pose estimate. This pose estimate is then published as a TF transform from base_footprint frame to odom frame. Once set up correctly, amcl will publish the odom to map transform, which will correct drift in the odometry and complete the base_footprint to map transform that the costmap node needs.

Next, the map layers. Below is an example YAML file for a costmap:

  global_frame: /map
  robot_base_frame: /base_footprint
  transform_tolerance: 0.2
  robot_radius: 0.6

  update_frequency: 5.0
  publish_frequency: 2.0
  rolling_window: false

  width: 100
  height: 100
  resolution: 0.05
  origin_x: -50
  origin_y: -50

  plugins:
    - {name: static, type: "costmap_2d::StaticLayer"}
    - {name: inflation, type: "costmap_2d::InflationLayer"}

  static:
    unknown_cost_value: -1
    lethal_cost_threshold: 100
    map_topic: /map

  inflation:
    inflation_radius: 1.1
    cost_scaling_factor: 0.2

In this example, there are two layer plugins: StaticLayer and InflationLayer. The static layer subscribes to the topic on the map_topic parameter, and puts lethal cost on cells that correspond to a value greater than lethal_cost_threshold in the loaded map.

The inflation layer adds cost to cells that are near lethal cost cells. This is done so that path planning algorithms that use the occupancy grid output are discouraged from planning extremely close to obstacles, but are still allowed to if absolutely necessary. The behavior of the inflation_radius and cost_scaling_factor parameters are nicely illustrated in the inflation layer plugin documentation linked above.

The robot_radius is also used to pad the costmap to avoid the footprint of the robot entering lethal cost cells. The effects of the robot's footprint on cost padding is also shown in the inflation layer documentation.

I hope this helps!