Revision history [back]

For the video you mention we didn't do any special filtering for reflective surfaces, but just let the system cope with the effects. Reflections are actually often reproducible and to the LIDAR/SLAM system things really look like there's "a room behind the mirror", with the "room" actually being used for scanmatching. It gets a little ugly if a robot tries to drive into this room of course (or if phantom obstacles are generated in otherwise free space). Actually detecting reflective/mirror surfaces reliably can be quite tricky and I don't know of a available open source solution for ROS for achieving that task. A classic advice would be "use another sensor with a different sensing modality" like a ultrasonic sensor. This can certainly help, but integrating that information with the rest of the software (SLAM as well as path planning) can be a non-trivial task.