we are coming from a ROS based system and are currently experimenting with isaac.
What we are doing is to test mapping, localization and navigation. We want to use the isaac cartographer wrapper for the mapping and localization.
Since we do have a robot that is based completely on ROS we wanted to test by just converting all sensor data to proto and use that in isaac.
What we’ve achieved so far is:
Cartographer does not work:
(I am not allowed to post more than one image) No map is shown and after closing the app , the map in /tmp is empty.
We are also confused by the documentation. According to this
cartographer does not need odometry as input as gmapping does.
However, according to this requirement cartographer needs some sort of odometry?
We have no idea what could be wrong with our cartographer app, since we use exactly the same app as for gmapping except including the cartographer subgraph and cartographer configs similar to the carter example.
We also cannot try the log_cartographerexample since the replay data is missing.
We use SDK version 2020.1.
Any help or pointers would be largely appreciated! Thanks in advance and kind regards,
I think i get it now. isaac::Cartographer reads odom directly from the PoseTree? isaac::GMapping let’s the user decide if it should be read from the PoseTree or from the odom RX.
If i set use_pose_tree=true for GMapping, it also does not work and mapping gets stuck after the first scan.
Could it be that reading the PoseTree does not work correctly for isaac::Cartographer as well, resulting in our posted problem?
If so, what could cause this behaviour? We only use RosToPose to transfer our ros::TF to isaac. Websight does show the correct transformations. So reading works at least for other components.
One way to do that is having a codelet that receives pose data from message channel and update the data into PoseTree periodically. Pose hook in engine/alice/hooks/pose_hook.hpp may come handy. You could define
ISAAC_POSE3(odom, sensor);
in the codelet and update data like
set_odom_T_sensor(pose_data, time_in_secs);
thank you very much for the reply.
We’ve setup a RosToPoses node thats maps all necessary tf poses to the PoseTree. The odom2robot transform gets correctly mapped from the tf::transformodom2base_link.
I can see that it works via websight. The poses are correctly synchronized to the PoseTree. However, GMapping is not working when GMapping::use_pose_tree settings is set to true. The laser scans are shown, but the map is not built, it seems that odom2robot is not read, or doesn’t change (although it does, as can be seen in websight::PoseTree).
My guess is, that this is the same prolem with cartographer since it directly reads from the PoseTree.
I tried to set the odom2robot manually as you suggested. I’ve put it into our ros::nav_msgs::Odometry to isaac::Odometry2Protoconverter, which is called at a rate of 50Hz. This does not work as intended though. After disabling our RosPoses mapper, the odom2robotshould now come only from our new converter node. If I std::cout the transform odom2robot it prints the correct values, but websight seems to be not synchronized. It seems to get not updated for longer periods. GMapping with use_pose_tree=true is also not working when using the manually published odom2robot transform. No warning or error is printed from isaac.
Is there a suggested rate for publishing PoseTree transformations? Could this be the reason of websight being not updated?
we are playing rosbags with old time stamps.
Could it be that the RosToPose mapper uses the old time stamps? Resulting in old transformations when GMapping or cartographer query the PoseTree with a current time as argument?
I’ll investigate that and will post my results.
edit, this is what i get if read the odom2robot using a custom codelet that does nothing except periodically reading the odom2robot from the pose tree:
2020-07-14 11:04:14.395 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 4.15523 -0.881271 rotation 0.797655
2020-07-14 11:04:14.495 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 0 0 rotation 0
2020-07-14 11:04:14.595 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 4.24858 -0.772422 rotation 0.938966
2020-07-14 11:04:14.695 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 0 0 rotation 0
2020-07-14 11:04:14.795 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 4.34188 -0.623418 rotation 1.08446
That seems strange. I use a pose initializer to set odom2robot to identity. I assume that this gets overwritten by RosToPoses?
Found out a a pose initializer does interfere. Removed the pose initializer and this happens:
2020-07-14 11:15:31.293 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 3.36078 9.62546 rotation 1.72432
2020-07-14 11:15:31.392 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 3.34834 9.70486 rotation 1.73326
2020-07-14 11:15:31.492 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 3.32959 9.81459 rotation 1.75306
2020-07-14 11:15:31.593 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 3.3176 9.87832 rotation 1.77064
2020-07-14 11:15:31.692 INFO packages/mojin_isaac/utils/PoseTreeTransformationReader.cpp@22: odom2robot: translation 3.30116 9.95658 rotation 1.79431
Edit:
I misunderstood the PoseInitializer. From the naming I concluded it is only an initializer, but it can be used to periodically publish a transform.
This can be controlled by params.
I’ve identified a major problem with how we use isaac and rely on played rosbags.
The PoseTree lives in another time as the messages.
Using sim_time=true and rosbag play --clock does not work since the RosToPoses still uses “current time”:
