challenge3

Challenge 3

Does Direct SLAM work better in an indoor area or outdoor area?

Approach

During the course of this challenge, we faced some system setup issues in installing packages for Direct SLAM, hence our experiments use LIDAR based Hector SLAM (branch: catkin).

However, from a Direct SLAM perspective, you may find these GitHub repos helpful. They make use of the popular LSD Slam package.

• https://github.com/IshitaTakeshi/lsd_slam_noros
• https://github.com/tum-vision/lsd_slam

Observations

Indoor:

• Hector SLAM works pretty well indoors with good detailing.
• The middle pillar in High Bay is also well detected.
• There are glass windows on one side which can be noticed as rays are not able to reflect back to the sensor.

Outdoor:

• As vehicle comes out of lab, gate is detected well. Vehicle localization is also good.
• Building boundary is detected well
• Then, due to open space, LIDAR sensor cannot capture back the reflected laser rays and the mapping and localization messes up.

Demo

Indoor SLAM

Outdoor SLAM

Setup

$ sudo apt-get install ros-kinetic-hector-slam

Make sure catkin is setup and a Catkin workspace is there. In my case, I call it, catkin_ws. For help, look at this.

Then, clone the above Hector Slam GitHub repo into catkin_ws/src and run catkin_make to build it.

I found this tutorial helpful for understanding how to use Catkin and work with Hector SLAM.

All Set!

Dataset

The ROS Bag files (indoor and outdoor) can be downloaded from here.

Running

No separate code files are required for completing this challenge. All required changes are done in already present internal files as detailed.

We basically follow the instructions in this tutorial. With no changes, you should be able to see the map being formed on the default tutorial rosbag file.

For us, we recorded the sensor data from the autonomous vehicle into a rosbag where the lidar data comes under the lidar1 topic. Hence, we will make some changes in the default tuturial run file.

Open catkin_ws/src/hector_slam/hector_mapping/launch/mapping_default.launch file and change the default value of arg name="base_frame" and arg name="odom_frame", both to lidar1.

Finally when running, map the lidar1/scan topic to /scan topic.

Indoor SLAM

rosbag play insidehighbay.bag /lidar1/scan:=/scan --clock -r 2.5

Note: The clock argument just runs the clock through the bag faster.

Outdoor SLAM

rosbag play outsidehighbay.bag /lidar1/scan:=/scan --clock -r 2.5