| ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange |
I'm not sure about this, but I gather that the slam_gmapping package cannot combine the laser scans from multiple robots?
If this is true, does anyone know if there exists a ROS slam package that does this? I've done a bit of searching but haven't been able to find one.
I am not sure about other SLAM packages that would support multi-robot mapping, but i can confirm, that GMapping is not capable of doing this.
In fact i would guess that none of the particle filter based SLAM variations (like GMapping) is suitable for this use case. Since the map you actually see is only the estimate produced by the currently highest rated particle there would be some kind of combinatory explosion of the needed particle count if odometry-noise from two or more robots has to be estimated. And since the initial offset of the starting points of the robots is not known, even more particles with guesses of this offset would be needed for a global map to align properly if one robot visits an area the other robot has already mapped.
You may have a look at graph-based approaches that maybe incorporate visual features that are globally unique (e.g. SIFT, SURF). But i really have no idea if there is a algorithm like this implemented in ros already. I would really appreciate if you could keep us posted about your findings later on.
No, slam_gmapping doesn't work with multirobot systems. I have tried it with stage, the SLAM just doesn't start to work, hopefully we will have some improvement in fuerte (March 2012 release).
Probably the reason that slam_gmapping fails is that the communication between two (or more) robots is ambigious.
This is a very old topic. But I have constructed a node that can take in laser and odom information from another robot and use that information to add it to it's own map. This is based on the principle of letting a virtual robot run back through the traversed path of the other robot. The implementation is still crude and obviously needs some polishing, but it works very well. You can either use a decentralized approach where the robots incorporate the data of all the other robots the entire time or in intervals (or upon encounter) or do it centralized and have 1 pc/or robot take in all the lasers and odom of the robots and build 1 map from it. What you essentially have to do is subscribe to the odom and laser of all the robots and republish them under the same namespace. You do have to keep track of the timings ofcourse, you need to restamp the message to the current timeframe so that GMapping doesn't reject them. It's a bit of work, but certainly doable.
I know it's been over a year, but this node would be very appreciated. I'm running into a similar problem and would love to see your code. Thanks
FYI, as part of the mrpt_graphslam_2d package, we have implemented an pose-graphSLAM variant that implements multi-robot graphSLAM. For more on this see:
Asked: 2012-02-25 18:53:40 -0500
Seen: 2,481 times
Last updated: Jun 29 '14
ROS Answers is licensed under Creative Commons Attribution 3.0 Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
