by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Randy Mackay
    @rmackay9
    @patrickpoirier51, ok, so you think maybe I need to move the vehicle around a bit before takeoff? lift it up and down etc? I did multiple flights without rebooting the camera and it doesn't seem that it figured out the scaling..
    pos-vs-vel-z.png
    Also here is an example where the camera's reported position and velocity are not matching. In general the velocity (in green) matches the position (in red) but at the moment with the yellow circle you can see the reported velocity is positive (meaning falling) but the position becomes more negative (meaning the vehicle is moving higher).. I guess this could be caused by high vibration in the camera's IMU.. my camera has no vibration isolation..
    Randy Mackay
    @rmackay9
    camera-glitch.png
    Also here is an example where the camera has apparently reset it's position but not told us.. we can see the camera's X and Y position has jumped but the VISP.reset_counter stays at "1" (the reset_counter isn't shown in the graph but it stays at "1"). I have another log where the reset-counter is always "2" so I know that the reset_counter can be higher than 1.. but in this case it doesn't seem to be working. Not sure if that's an issue in the scripts or the camera itself..
    The bentwing 3D Aerial Mapping System
    sorry for interrupting the discussion
    patrickpoirier51
    @patrickpoirier51
    @thien did you had a chance to test your scripts on actual flights yet?
    @rmackay9 yes I experienced the reset counter but it made the EKF stop working so the vehicle couldn't arm
    patrickpoirier51
    @patrickpoirier51
    Oh you are running BOTH position and velocity ?
    Randy Mackay
    @rmackay9
    @patrickpoirier51, yes, both position and velocity. the reset counter shouldn't cause the EKF to stop working anymore at least with EKF3. EKF3 has glitch protection and position reset handling. EKF2 only has position reset handling and the glitch protection still doesn't work and can cause the EKF to get into a bad state that it won't recover from.. basically what happens is the EKF2 recognises the glitch but there is an error in the position reset code so it starts endlesslying doing position resets..
    Randy Mackay
    @rmackay9
    I just did another flight after first confirming that the scaling was OK indoors.. it still didn't work very well and again I saw a position reset (of a few meters) with no update of the reset_counter..
    I think we will probably need to implement a pre-arm check for the T265 backend so that we can't takeoff in Loiter (etc) unless the camera is reporting "High Confidence"..
    Thien Nguyen
    @thien94
    @patrickpoirier51 Unfortunately I can't do any flight tests at the moment. Hopefully in a few weeks time I will be able to.
    @rmackay9 it seems there are many intertwined problems at the moment. I have updated the list here https://github.com/thien94/vision_to_mavros/issues/15#issue-593727565, we can work on it one by one.
    Thien Nguyen
    @thien94
    The vibration would definitely affect the IMU -> velocity estimate -> position estimate. The reset_counter part definitely needs more work. Right now on my T265 I can say the reported value does not reflect all of the jumped cases, so maybe we need to impose an additional check by ourselves.
    Randy Mackay
    @rmackay9
    @thien94, nice list. I've crossed linked my AP code list and your list. Txs for the info on the camera's problems with reporting resets. I wonder if we could raise an issue on the Intel github site.. or will they just close it :-). I agree with you that if we can't get the camera to correct report resets we will need to implement something on our side.
    @anbello, thanks for the logs. Interestingly your log also shows a scaling issue vs the baro but you also have a rangefinder which matches the camera almost perfectly meaning it's the baro that's the issue. I guess I'll attach a rangefinder as well.
    image.png
    Randy Mackay
    @rmackay9
    @anbello, your position and velocity innovations are far better than mine. Here's a couple of graphs, your log is on the top, mine is on the bottom.
    pos-innovations.png
    vel-innovations.png
    So your flights are a success I think, thanks very much!
    @anbello, I guess your flights are indoors? I think I will try flying indoors next time. Another difference is I have the compass enabled while you are taking yaw from the camera... I will try this setup as well to see if it makes a difference.
    Randy Mackay
    @rmackay9
    @patrickpoirier51, you came up with a 3D printed mount right? Did that include vibration dampening? Maybe we should try and retrieve the IMU values from the camera and detect if vibes are too high?
    Andrea Belloni
    @anbello
    @rmackay9 thanks for analyzing my logs, and yes they are both from indoor flights.
    I saw issue with baro, maybe temperature compensation may help, I have to try with TCAL_ENABLED = 2, what do you think about this?
    Randy Mackay
    @rmackay9
    @anbello, I've never personally used the baro tcal but I don't think it will help. It's probably ground effect.. the vehicle is only climbing a meter or two in the test and it's possible that it would look quite different if the vehicle flew higher. I actually don't think we need to worry about it for now cuz baros are pretty reliable and it's not causing any performance issues.
    @thien94, despite my various troubles today I think the next step is still to try and get the confidence info flowing into AP. It's a Groups decision of course..
    patrickpoirier51
    @patrickpoirier51
    @rmackay9 I will mount it on a damping platform and test
    Andrea Belloni
    @anbello
    @rmackay9 the T265 works better indoor because there are many more features to track (usually) as pointed also from @patrickpoirier51
    for example this is the room where I do my test
    https://youtu.be/lUMYLjNW7n4
    Andy Piper
    @andyp1per
    @anbello do you use the FFT notch in your test?
    Andrea Belloni
    @anbello
    @andyp1per of course :)
    Andy Piper
    @andyp1per
    I am psyched! That's clearly what @rmackay9 is missing ;)
    Andrea Belloni
    @anbello
    @rmackay9 was talking about dampening vibration of the IMU in the T265
    Andrea Belloni
    @anbello
    T265 vibration affect negatively both IMU measurements and visual features tracking, IMU measurements can be SW filtered but for the other problem I think we could only do HW damping
    Randy Mackay
    @rmackay9
    I re-tested the changes from Chobitsfan and I today in my garage and it worked a-ok. So I used EKF3 with position and speed estimates. The height was also taken from the camera and I did a few tests, some with compass as the heading source, some with the camera as the heading source and they performed equally well.
    So it's pretty clear that my issues yesterday were just the outdoor environment causing the camera troubles
    @anbello, your room has tons of features! I can imagine the camera is very happy in this environment. I might post an unlisted video later of my much more bleak garage..
    Randy Mackay
    @rmackay9
    Here's the video from my garage: https://youtu.be/jCY9-fWtA7c
    Andrea Belloni
    @anbello
    Also in my test indoor was always better then outdoor, I have yet to try outdoor with 45° tilted t265, maybe there are more chance to have good features
    patrickpoirier51
    @patrickpoirier51
    @rmackay9 I made some features with duct tape https://youtu.be/aQcRiFnSpGA
    @anbello the 45deg down might induce a tendency to make the vehicle move forward
    Thien Nguyen
    @thien94
    Hi @rmackay9 , since the vision_speed_estimate seems to work ok, I have added the following new updates in the same branch: https://github.com/thien94/vision_to_mavros/tree/t265_noti_event
    • covariance of position, attitude, speed from confidence level: first attempt following this code from intel.
    • a manual check based on the displacement of the position for pose jump event (along with the API provided callback), albeit a simple one but it seems to work predictably in my handheld tests.
    Regarding the pose jump, one thing I have noticed in my tests is that the pose jump happens more often a while after a period of hectic movements, when the vehicle has settled down.
    Randy Mackay
    @rmackay9
    @patrickpoirier51, nice! I thought about doing that but it did ok in any case.
    Randy Mackay
    @rmackay9
    @thien94, thanks very much for the confidence-level code. I'll add the AP side this week..
    chobitsfan
    @chobitsfan
    Hi @rmackay9 Could you be able to take a look at @thien94 's comment https://github.com/ArduPilot/ardupilot/pull/14368#discussion_r429187408 ? Thank you very much
    Randy Mackay
    @rmackay9
    @chobitsfan, @thien94, OK, np. We can remove the scaling from the velocity.. I'll add commit..