Assets writer

The purpose of SLAM is to compute the trajectory of a single sensor through a metric space. On a higher level, the input of SLAM is sensor data, its output is the best estimate of the trajectory up to this point in time. To be real-time and efficient, Cartographer throws most of the sensor data away immediately.

The trajectory alone is rarely of interest. But once the best trajectory is established, the full sensor data can be used to derive and visualize information about its surroundings.

Cartographer provides the assets writer for this. Its inputs are

  1. the original sensor data fed to SLAM in a ROS bag file,
  2. the cartographer state, which is contained in the .pbstream file that SLAM creates,
  3. the sensor extrinsics (i.e. TF data from the bag or a URDF),
  4. and a pipeline configuration, which is defined in a .lua file.

The assets writer runs through the sensor data in batches with a known trajectory. It can be used to color, filter and export SLAM point cloud data in a variety of formats. For more information on what the asset writer can be used for, refer to the examples below below and the header files in cartographer/io.

Sample usage

# Download the 3D backpack example bag.
wget -P ~/Downloads https://storage.googleapis.com/cartographer-public-data/bags/backpack_3d/b3-2016-04-05-14-14-00.bag

# Launch the 3D offline demo.
roslaunch cartographer_ros offline_backpack_3d.launch bag_filenames:=${HOME}/Downloads/b3-2016-04-05-14-14-00.bag

Watch the output on the commandline until the offline node terminates. It will have written b3-2016-04-05-14-14-00.bag.pbstream which represents the Cartographer state after it processed all data and finished all optimizations. You could have gotten the same state data by running the online node and calling:

# Finish the first trajectory. No further data will be accepted on it.
rosservice call /finish_trajectory 0

# Ask Cartographer to serialize its current state.
rosservice call /write_state ${HOME}/Downloads/b3-2016-04-05-14-14-00.bag.pbstream

Now we run the assets writer with the sample configuration file for the 3D backpack:

roslaunch cartographer_ros assets_writer_backpack_3d.launch \
   bag_filenames:=${HOME}/Downloads/b3-2016-04-05-14-14-00.bag \
   pose_graph_filename:=${HOME}/Downloads/b3-2016-04-05-14-14-00.bag.pbstream

At the time of writing, the generated assets end up in ~/.ros.

Configuration

The assets writer is modeled as a pipeline. It consists of PointsProcessors and PointsBatchs flow through it. Data flows from the first processor to the next, each has the chance to modify the PointsBatch before passing it on.

For example the assets_writer_backpack_3d.lua uses min_max_range_filter to remove points that are either too close or too far from the sensor. After this, it writes X-Rays, then recolors the PointsBatchs depending on the sensor frame ids and writes another set of X-Rays using these new colors.

The individual PointsProcessors are all in the cartographer/io sub-directory and documented in their individual header files.

First-person visualization of point clouds

Generating a fly through of points is a two step approach: First, write a PLY file with the points you want to visualize, then use point_cloud_viewer.

The first step is usually accomplished by using IntensityToColorPointsProcessor to give the points a non-white color, then writing them to a PLY using PlyWritingPointsProcessor. An example is in assets_writer_backpack_2d.lua.

Once you have the PLY, follow the README of point_cloud_viewer to generate an on-disk octree data structure which can be viewed by one of the viewers in the same repo.