Troubleshooting and bugs - OwlSoul/tango_sensors GitHub Wiki

Known issues

Application hangs, especially point cloud

Usually happens on low battery (close to 0%). Avoid using this app on a low battery level, it will become unstable (due to Tango parts and network being unstable at low battery level). If your tablet is old, or charging, battery level may display wrong charge level, and as soon as you unplug the power it will shut up immediately due to no power left. This is normal.

No connection to ROS issue

App will not tell you if it didn't connect to ROS Master or not, use ROS tools to check this. Technically this is a feature (we needed this to be able to use the app even if ROS Master is down). Also, if ROS Master will become unreachable, app will not automatically reconnect. App cannot be it's own ROS master as well. Check if tablet is seen and messages arrive on the ROS side to ensure everything works properly.

Point cloud is shaking when tablet turns, not smooth like in "ros_tango_streamer"

When point cloud is published, a corresponding transform which equals the pose a the moment point cloud is available being published. No Iterative Closest Point filtering or similar algorithms are implemented, data is raw. You may ignore transforms completely and build your own processing algorithm in ROS, tablet does noting anyway, and ALL transforms published are for demonstration purposes only. App does not align point_cloud with world coordinates.

Publishing rate gradually reduced

Try to avoid using this app with WiFi connection if you need maximum publish rate, using USB tethering is the best option. Publishing color camera images at max resolution will "eat" even USB tethering connection, and don't forget and underestimate Point Cloud data, which may take more than 5 MB/s when it detects a lot of points. Reduce publishing rate in the app when possible by setting a delay after send (can be done for all Tango-related sensors).

Notes about Pose estimation

No area learning or drift correction are being implemented. App is designed to provide raw data only, no Tango SLAM capabilities included.

Point cloud transformation issue

If you point the tablet to a wall while it is completely parallel to such wall, and the plane formed by point cloud is not perpendicular to the pose. In Yellowstone tablet, camera is being tilted compared to back tablet surface, so when you're pointing your tablet to a wall, camera is not perpendicular to it. Since this app supplies raw data with some transformations, you can switch on point cloud and pose estimation, and look in RVIZ when the "wall" plane will be perpendicular to the pose vector, this is when your camera, not tablet, is pointing to a wall perpendicularly. Do not forget about this issue when you're working with Tango.

Tango is "looking" at the wall

Note that the "plane" formed by points is perpendicular to the pose estimation, while pose estimation points slightly down. This is the position when your camera is looking directly at the wall.

Fisheye camera is not working

There's a chance Fisheye camera will not work due to API version. According to tango_ros_streamer page, they state that "Camera info of the Tango device fisheye camera. The fisheye camera data is not available on devices with API level > 23. This topic is therefore not advertised for these devices". This is not tested with "tango_sensors" app, and never encountered on two Yellowstone tablets with Kitkat Android used in testing.