Sensors and datasets

This page lists the possibilities to feed SLAM algorithms with sensory data in the MOLA framework.

1) Off-line: read from datasets

Parsing data from a dataset is the easiest way to test, debug, and benchmark any kind of localization, mapping, or classification algorithm. Existing options are:

From a ROS bag

Using mola-input-ros1 as a bridge between ROS->MOLA, then replaying the ROS bag using standard ROS tools, i.e. rosbag play.

From an MRPT rawlog

Using mola-input-rawlog to parse a .rawlog file and have all its observations delivered to the rest of MOLA modules (sensor front-ends).

From public robotics datasets

Specific MOLA modules exist to directly parsing popular public datasets:

  • mola-input-euroc-dataset

  • mola-input-kaist-dataset

  • mola-input-kitti-dataset

2) On-line: direct sensor connection

There exist two alternatives:

  • to use ROS drivers for sensors and then use mola-input-ros1 as a bridge between ROS->MOLA.

  • to directly connect to any sensor supported by mrpt-hwdrivers using the MOLA native mola-input-hwdrivers.

In general, using mrpt-hwdrivers should be more efficient and convenient. A non-exhaustive list of sensors supported by this library is:

  • 2D lidars: Hokuyo, SICK, RP-Lidar, Ibeo.

  • 3D lidars: Velodyne.

  • Cameras: Any camera supported by OpenCV and ffmpeg. Bumblebee2 stereo cameras.

  • Depth sensors: Original XBox Kinect, any camera using OpenNI2.

  • IMUs: xSens, KVH DSP3000.

  • GNSS (GPS): Any sensor providing a standard NMEA interface. Novatel receivers.