Robotics StackExchange | Archived questions

Rtabmap localization with rgbd and lidar

I have a 2d lidar and rgbd camera. The camera is located very low, like 5 cm above the ground and I have few questions about the rtabmap

  1. My environment is very dynamic and very often the concrete floor has a lot of features. Does rtabmap localized based on the features? If yes, how good is it?

  2. Does rtabmap fuses both lidar and camera localization? As I saw that rtabmap has ICP, is it only for the odometer or for the pose estimate?

Asked by EdwardNur on 2019-10-18 14:01:16 UTC

Comments

Answers

  1. By features, do you mean visual or geometric features? rtabmap uses visual features for global localization and can use lidar to refine localization. For loop closure detection / localization comparison with other algorithms, see this paper (Appearance-Based Loop Closure Detection for Online Large-Scale and Long-Term Operation).

  2. ICP can be used to refine odometry and to refine loop closure detections / localizations. For example, the motion estimation of a loop closure / localization is computed visually from the camera, then optionally refined using ICP with the lidar. The whole pipeline is described in this paper (RTAB-Map as an Open-Source Lidar and VisualSLAM Library for Large-Scale and Long-Term OnlineOperation).

cheers,
Mathieu

Asked by matlabbe on 2019-10-22 16:37:35 UTC

Comments

@matlabbe I see. So if I want to localize both with visual and ICP, should I choose Reg/Strategy 2?

Asked by EdwardNur on 2019-10-24 14:40:06 UTC

Reg/Strategy=1 is more efficient. If needed (in case we don't have any a priori guess of the transformation), a visual estimation will still be done using camera.

Asked by matlabbe on 2019-11-03 15:31:23 UTC