I'm extensively testing T265 and its v-slam.
The testing is done on a floor of 20m x 30m with several rooms. Height changes are ignored. T265 is attached to the top of a helmet, shooting towards ceiling during the test. An operator takes this helmet and walk around the floor.
The example ar-advanced from "realsense SDK" is modified so that multiple static nodes can be added by key pressing during the test. 7 landmarks (rs2::set_static_node) are marked out with hopping that they will be used for re-localization and consequently long-term drift will be minimized.
However, as shown in the test results, two rooms are drift away in about 5 minutes walking, re-localization never happened during the test.
Could someone help to check if I have blind points (mistake) during this test?
Can T265 do real slam with map optimization at the loop point?
Please sign in to leave a comment.