cartographer_ros
cartographer_ros copied to clipboard
cartographer-project
cartographer_ros and Intel Research Lab dataset leads to strange behaviour
Hello, I need to use Intel Research Lab dataset (available at here) in cartographer_ros package (ROS Melodic). The dataset is in carmen-old log format, therefore it is converted to rosbag file using carmen_publisher node (see here). When I run cartographer_ros with generated bagfile, however, trajectory goes everywhere and fails to generate map. I tried every possible things I can come up with to fix this. I guess this is related to wrong tf configurations, but I cannot figure out what to do.
Looking at tf tree and node graph, everything seems fine, but actually it is not.
- results for
cartographer_rosbag_validateI rancarmen_publishernode for generating bagfile:
$ rosrun carmen_publisher carmen2rosbag.py -i intel.log -o intel.bag
[INFO] [1573457171.581068, 0.000000]: Reading data from : intel.log
[INFO] [1573457171.582209, 0.000000]: Writing rosbag to : intel.bag
[INFO] [1573457171.583196, 0.000000]: Start convertion to rosbag
[INFO] [1573457171.584270, 0.000000]: Param 'robot_frontlaser_offset' has value : 0.0
[INFO] [1573457171.585185, 0.000000]: Param 'robot_rearlaser_offset' has value : 0.0
[INFO] [1573457177.348353, 0.000000]: Job's done.
$ rosbag info intel.bag
path: intel.bag
version: 2.0
duration: 44:51s (2691s)
start: Dec 06 2000 06:47:37.34 (976052857.34)
end: Dec 06 2000 07:32:28.63 (976055548.63)
size: 34.6 MB
messages: 67461
compression: none [46/46 chunks]
types: nav_msgs/Odometry [cd5e73d190d741a2f92e81eda573aca7]
sensor_msgs/LaserScan [90c7ef2dc6895d81024acba2ac42f369]
tf2_msgs/TFMessage [94810edda583a504dfda3829e70d7eec]
topics: /FLASER 13631 msgs : sensor_msgs/LaserScan
/ODOM 26915 msgs : nav_msgs/Odometry
/tf 26915 msgs : tf2_msgs/TFMessage
then I ran cartographer_rosbag_validate command and got the result as shown below:
$ cartographer_rosbag_validate -bag_filename intel.bag
E1111 16:28:00.829424 28509 rosbag_validate_main.cc:382] Point data (frame_id: "base_link") has a large gap, largest is 3.00266 s, recommended is [0.0005, 0.05] s with no jitter.
I1111 16:28:00.829820 28509 rosbag_validate_main.cc:398] Time delta histogram for consecutive messages on topic "/FLASER" (frame_id: "base_link"):
Count: 13630 Min: 0.000006 Max: 3.002663 Mean: 0.197456
[0.000006, 0.300272) ############### Count: 9988 (73.279533%) Total: 9988 (73.279533%)
[0.300272, 0.600537) #### Count: 2998 (21.995598%) Total: 12986 (95.275131%)
[0.600537, 0.900803) # Count: 388 (2.846662%) Total: 13374 (98.121788%)
[0.900803, 1.201069) Count: 74 (0.542920%) Total: 13448 (98.664711%)
[1.201069, 1.501335) Count: 53 (0.388848%) Total: 13501 (99.053558%)
[1.501335, 1.801600) Count: 47 (0.344828%) Total: 13548 (99.398384%)
[1.801600, 2.101866) Count: 34 (0.249450%) Total: 13582 (99.647835%)
[2.101866, 2.402132) Count: 38 (0.278797%) Total: 13620 (99.926636%)
[2.402132, 2.702398) Count: 8 (0.058694%) Total: 13628 (99.985329%)
[2.702398, 3.002663] Count: 2 (0.014674%) Total: 13630 (100.000000%)
I1111 16:28:00.830135 28509 rosbag_validate_main.cc:398] Time delta histogram for consecutive messages on topic "/ODOM" (frame_id: "odom"):
Count: 26914 Min: 0.000034 Max: 3.107075 Mean: 0.099989
[0.000034, 0.310738) ################# Count: 23215 (86.256226%) Total: 23215 (86.256226%)
[0.310738, 0.621442) ## Count: 2910 (10.812217%) Total: 26125 (97.068443%)
[0.621442, 0.932146) Count: 488 (1.813183%) Total: 26613 (98.881622%)
[0.932146, 1.242850) Count: 110 (0.408709%) Total: 26723 (99.290329%)
[1.242850, 1.553554) Count: 57 (0.211786%) Total: 26780 (99.502121%)
[1.553554, 1.864259) Count: 48 (0.178346%) Total: 26828 (99.680466%)
[1.864259, 2.174963) Count: 41 (0.152337%) Total: 26869 (99.832802%)
[2.174963, 2.485667) Count: 36 (0.133759%) Total: 26905 (99.966560%)
[2.485667, 2.796371) Count: 7 (0.026009%) Total: 26912 (99.992569%)
[2.796371, 3.107075] Count: 2 (0.007431%) Total: 26914 (100.000000%)
-
Intel research lab dataset I uploaded several files for reproducing the problem to GitHub Gist: https://gist.github.com/sterngerlach/efc7de1f5cc1da16f2feeb1562aaa1e0
-
Carmen log file and generated bagfile I uploaded original log file and generated bagfile to Google Drive: https://drive.google.com/drive/folders/1R1HKzrAi4IEtjRA8lAuRv3wnLK751R-3?usp=sharing
In fact, this bagfile works great with slam_gmapping package. I also uploaded files necessary for running slam_gmapping to GitHub Gist: https://gist.github.com/sterngerlach/b8aac3c6d6581537cd248471af3c5e1f
I hope someone could help me to fix this problem. Thank you.
@sterngerlach I meet the same problem with you. What's more, in my case the topic "tf" in the rosbag converted by Intel.log does not show laser link.
I refer to the setting of FabianSchurig and use a high weight to lock the drift of tf, but I don’t think this is the right approach.
There were even four times of huge tf reversal that it could not be controlled, so I reset the trajectory.
I still don’t understand why the laser data causes such a huge jump in tf .
https://gist.github.com/eithwa/099d119137c2a4d982ec82c2e3d660a9
I get good result with old version cartographer ceres-solver cartographer cartographer_ros configure
The odometry error is too large, so mapping failed. The odometry must be corrected to work properly.
The odometry error is too large, so mapping failed. The odometry must be corrected to work properly.
May I ask your parameter configuration? I tried not to use odom but it did not help to construct the map
The odometry error is too large, so mapping failed. The odometry must be corrected to work properly.
May I ask your parameter configuration? I tried not to use odom but it did not help to construct the map
I modified the source code of cartograher, turn off laser scan motion undistortion, and make sure scan and odom are synchronized.
The odometry error is too large, so mapping failed. The odometry must be corrected to work properly.
May I ask your parameter configuration? I tried not to use odom but it did not help to construct the map
The map cannot be created correctly by modifying the parameters.
Turn off laser scan motion undistortion
The map you built is very clear! Can you share your code, I want to know what changes you made. Thank you very much!
The odometry error is too large, so mapping failed. The odometry must be corrected to work properly.
May I ask your parameter configuration? I tried not to use odom but it did not help to construct the map
I modified the source code of cartograher, turn off laser scan motion undistortion, and make sure scan and odom are synchronized.
The map you built is very good, could you please share the intel lab dataset in rosbag format and configuration files you used? Thank you.
