Rosanswers logo

I want to do mapping with the kinect, so my robot can avoid obstacles and move around. Looking around there seems to be several options, but not sure which is better (better = low computational requirements):

  1. rgbdslam has a lot of interest and builds cool looking 3D world, but I believe it runs slow (2 fps) and requires a lot of computational power.

  2. gmapping seems to do what I want, but is designed for laser rangers and not the kinect, so I am not sure if the overhead of converting from kinect->laser_scans now makes it perform as number 1 above.

  3. ethzasl_mapping seems to do a similar job to rgbslam (and it is also designed for the kinect) but there is no mention of performance.

I think these are the predominate packages for mapping. Would appreciate any thoughts on these or suggestions of what I am missing.

Unfortunately I am also have difficulty getting any of these to compile on Fuerte for OSX.


Originally posted by Kevin on ROS Answers with karma: 2962 on 2012-05-04

Post score: 4


1 Answer 1


Rosanswers logo

I can't tell you what is best, but you can speed up rgbdslam significantly by

  • decreasing the feature count parameters
  • decreasing the "geodesic_candidates" and "min_sampled_candidates"
  • setting the kinect driver to QVGA (e.g., using dynamic_reconfigure)
  • using SIFTGPU features, if you have a GPU, ORB otherwise
  • compile in "Release" configuration

However, the first three options and using ORB will decrease the accuracy.

Originally posted by Felix Endres with karma: 6468 on 2012-05-05

This answer was ACCEPTED on the original site

Post score: 6


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.