Home

Welcome to the strands_karl wiki!

Here are some todos we need to work on. For each of them we already have a header with some pointers showing you where to look for info or what to try first. If you got the info from some other site, also put a link there, so we know where to check if something stops working. Please to try do them in the listed order, as this list is sorted based on easiness and importance.

Starting the robot

Start up the robot, like for the marathon, but without the reporter.

• Update the system with sudo apt-get update and sudo apt-get dist-upgrade
• Run new tmux file: ~/overlay/src/strands_karl/scripts/karl_bringup.bash will be here (look here for original)

People Perception

Get people perception running. I.e. make sure you can see the output of the upper body detector and the output of the tracker.

• Run: roslaunch perception_people_launch people_tracker_robot.launch load_params_from_file:=true which will launch multiple nodes from strands_perception_people. Note that launching just the upper_body_detector does not seem to do the job, i.e. the node is running and advertising the detection topic, but not broadcasting any messages.
• To show the upper body detector: run rviz and display the PointCloud2 from head xtion and the /upper_body_detector/marker_array topic which contains the detector output. In Rviz, the tracker output is visualized in the form of green cube around the person's shoulders.
• The tracker output is advertised on /people_tracker/marker_array. Smooth.

Debug No-go map problems

Debug the no-go map problem. Apparently we cannot have a nogo map and a normal costmap at the same time, check also the original issue.

Follow navigation issue and report

Keep track of strands_movebase#27 and strands_movebase#30 and see if something comes up:

• strands_movebase#30 is an update that contained updated params for strands_movebase. We tested the new params by having Carl go to the FreedomLookout (where he always used to fail) and he got there! The pull request was therefore merged.

Data playback from mongodb

Playback data from the mongodb, right now there are video streams and 3d meta room stored. We want to retrieve this data so we can work with.

• How to play back the video:

Run rosrun mongodb_store mongodb_play.py /head_xtion_compressed_rgb_compressed /head_xtion_compressed_depth_libav to play all the entries from the mongodb store, chronologically ordered. You can also use parameters specifying the start and end Datetimes of the playback like so: rosrun mongodb_store mongodb_play.py -s 21/11/14 22:30 -e 21/11/14 22:32 /head_xtion_compressed_rgb_compressed /head_xtion_compressed_depth_libav. To be able to visialize the depth image in Rviz, run rosrun image_transport republish libav in:=/head_xtion_compressed_depth _image_transport:=libav raw out:=/head_xtion/depth/image_raw. To see the video and depth map in Rviz, open it up and add two 'Image' blocks and have them listen to the two topics.

If you want to see on what dates there are some entries in the mongodb, check out the roslog database with the mongo shell and run db.head_xtion_compressed_depth_libav.find({ "_meta.inserted_at": { $gte: new ISODate("2014-11-28T22:13:01.31Z") }}, {"_meta.inserted_at": 1})

• How to get the pcl pointclouds from the meta room, both intermediate and complete clouds. (Check here for initial info.)

Data replication

Data replication from Karl to our Server. (For this we need to setup the server.) All the things are stored in a mongodb on the main pc, but this is (and will always again be) full so we need to move the data to another server.

Setup a recording routine

• Clear up space from the robot by dumping the database.
• Create a new map. (if needed)
• Setup waypoints at useful positions. (We will talk about this.)
• Create a task that will go to (some of) the waypoints and records some information about humans there, as in a video stream, including the output of the upper body detector and the tracker, with time and location.

Output Tracking bounding boxes in image space

Currently the bounding boxes are published in 3D coordinates. Some people need them in 2D image coordinates. This info is being computed already, but it needs to be published to a topic as well. Check out this.