Hey Joe, some tips for your Kalman filter from a controls guy: 1) It looks like you assume 0 uncertainty in your GPS measurements, since your covariance drops to 0 at the GPS updates. Giving this a plausible value will help smooth out the jumps you see in the velocity and position plots. This also becomes pretty important at low speeds where GPS uncertainty is larger compared to your motion. An easy way to find out this info is to either look at the specs for your receiver, or take a log of the GPS speeds and positions at rest and calculate the standard deviation of that data. Note that this may be dependent on GPS signal quality, which your chip probably provides a real-time measure of, but that’s probably not going to matter _too_ much for your purposes. 2) You mentioned at 8:52 that you think you have an error in the covariance plots since x, y, and z axes are identical. But this is actually expected if the estimated process and measurement noise on each axis is identical, the state transition equations are symmetrical (which they should be), and of course your initial covariance estimate is the same. If you look at the the Kalman filter equation to update your state estimate covariance matrix P, you'll see that it is not dependent on either the state or your measurements. So your covariance will be deterministic given a known update rate, and so should only not be identical across axes if you (a) have different noise models for each axis, (b) have additional sensors on one or more axes, or (c) have updates at different times for each axis (which doesn't look like the case). The 'sawtooth' pattern is characteristic of a blended IMU-GPS pair, so that should give you more confidence that you did it right! 3) And as a last note, don't forget to give your covariance variables units too. 🙂 Personally I find it conceptually helpful when looking at variance data to convert them to 3sigma values instead. This means you’d be looking at uncertainty in units of velocity rather than a variance “area” with units velocity squared. 3sig = 3*sqrt(var)
Agreed with the changes; I was like wait you defiantly should not have your GPS be trusted that much. Shouldn't the variance climb to a rough steady state with some oscillation that varies in amplitude based upon how closely the GPS and IMU agree. I know on our bots when they do rapid direction changes the variance goes way up because the lag from GPS means its always predicting motion in the wrong direction but in a straight line it drops a lot both because GPS is more accurate under those conditions but also because the inertial data matches the GPS track.
he should add a heater and an insulating enclosure. It will decrease the drift in sensor readings and thus reduce inaccuracy both within a flight and between flights.
At 8:40, I had absolutely no idea what was coming, and I didn't even know what it was when you said it. In fact, I didn't understand most of this video, but it's still entertaining enough for me to watch the whole way through.
Technician: ... um ... Flight ... we're getting a slight oscillation in the 'Z' axis. Flight Boss: ... hu? ... Oh ... it's ok, that's just J.B.'s signature "Z axis wiggle". Technician: (puzzled frown) you mean he designed it to do that? Flight Boss: ... um ... ... yes ... yes he did ... now pay attention to your board.
Congratulations! I watched the entire AVA assembly to this point (many of your other videos as well). Your persistence is contagious. It helps keep me motivated for my own projects. Hopefully, in the near future, they'll be a legitimate sustainable business. As practical and frugal as you are, I really appreciate the effort you put into aesthetic details, from your video editing to your PCB design.
"Let's say it at the same time, I know what you're thinking, ready, ok, it's" Me: "Another cool graph" Joe: "The process covariance estimates" Me: "Yea, that's exactly what I said"
Fantastic flight! Thanks for not only sharing the video of the rocket, but also breaking down the data in such an easily understandable way. I showed this to my wife (1st time viewer) and she was able to follow along just fine, which is a real testament to just how well you explain things. Good job!
Yes, it’s joey again! Love the rapid launch and video schedule due to the current... *pandemic* Planning on building my own rocket, inspired by you! The reaction control tutorials are very helpful You have definitely been an inspiration to many people :) I also plan to make a liquid fuelled rocket after the practice with homemade (very large) sugar motors.
Fantastic as usual, Joe. Really enjoyed the detailed information about the operation of the flight. As an aside, since I'm working on some pretty over-equipped environmental sensors right now and have sensors on the brain: have you considered the use of magnetometers to aid in orientation sensing? They're available in very small footprints (e.g. NXP MAG3110, which is 2x2mm DFN10) and in my experience they offer pretty decent measurements at nearly 80Hz, which might help you improve your integrator accuracy during alignment correction, and better sense pitch during flight. The noise isn't the best on the MAG3110, at about 0.3uT in practice (so 0.6% error on a standardised 50uT earth field). That translates to about 0.55 degrees of error on an axis if you're normalising to 90 degrees per axis, but you can get more expensive alternatives like the RM3100 that offer incredible resolution and noise floors in the order of tens of nanoteslas. I figure if you've already built the Kalman filter it shouldn't be too much work to add geomagnetic orientation values into the mix and keep the covariances low. Anyway, enough back-seat engineering from me. Good flight and I look forward to future videos :)
One of the things that I like about this is someone could find this some time in the future and it will still be useful not just a cool launch but the data to go with it. *So cool!!!*
On the topic of servos, the RC helicopter world has several reasonably priced strong lightweight high speed brushless servo manufacturers. I specifically bring them up because the servos on the heli swash plate have to have a consistent response speed. or the models become uncontrollable. BK servos in particular are spectacular.
Sir,great video as always,A AM A NOOB AT THIS STUFF,but here is an idea, maybe use a high res quarature rotary encoder to characterize the servo(or make a lookup table) and use the pot of the servo with an external high speed ADC with a different micro controller for total control over that TVC. Man That KF looks so much better. Best of luck !!! TH-cam lacks content of this caliber and quality and fun and SO MUCH FUN WOW.
Hi Joe, Have you thought about using your camera data to map the flight path in a 3d program? I work in CGI so can point you in the right direction and help trouble shoot any problems you may have if you want to go down that route. Cant wait for launch 9
Incredible stuff, I love your channel. Would it be possible to determine the wind velocity versus altitude profile during the first half of the flight as you fly? Could that information then be used to target a trajectory that would have the parachutes drift you back to the launch pad?
dude which provide more stability tvc or fin control like in missle in model rocket. can u make more projects on fin contact as it would be better for people since they have cheap rockets and servo need not take that much stress
Great to see a successful Sprint 8 flight. Looking forward to Sprint 8 and 9. Achieving 1Km. altitude will be amazing. What is your goal after the 1Km flight?
Really liking the real-time speed of the playbacks instead of the slow mo!
yes
No you don’t
Hey Joe, some tips for your Kalman filter from a controls guy:
1) It looks like you assume 0 uncertainty in your GPS measurements, since your covariance drops to 0 at the GPS updates. Giving this a plausible value will help smooth out the jumps you see in the velocity and position plots. This also becomes pretty important at low speeds where GPS uncertainty is larger compared to your motion. An easy way to find out this info is to either look at the specs for your receiver, or take a log of the GPS speeds and positions at rest and calculate the standard deviation of that data. Note that this may be dependent on GPS signal quality, which your chip probably provides a real-time measure of, but that’s probably not going to matter _too_ much for your purposes.
2) You mentioned at 8:52 that you think you have an error in the covariance plots since x, y, and z axes are identical. But this is actually expected if the estimated process and measurement noise on each axis is identical, the state transition equations are symmetrical (which they should be), and of course your initial covariance estimate is the same. If you look at the the Kalman filter equation to update your state estimate covariance matrix P, you'll see that it is not dependent on either the state or your measurements. So your covariance will be deterministic given a known update rate, and so should only not be identical across axes if you (a) have different noise models for each axis, (b) have additional sensors on one or more axes, or (c) have updates at different times for each axis (which doesn't look like the case). The 'sawtooth' pattern is characteristic of a blended IMU-GPS pair, so that should give you more confidence that you did it right!
3) And as a last note, don't forget to give your covariance variables units too. 🙂 Personally I find it conceptually helpful when looking at variance data to convert them to 3sigma values instead. This means you’d be looking at uncertainty in units of velocity rather than a variance “area” with units velocity squared. 3sig = 3*sqrt(var)
I have no idea what you just said, but it sounds like he should do that
@@silience_ Can relate.
Agreed with the changes; I was like wait you defiantly should not have your GPS be trusted that much.
Shouldn't the variance climb to a rough steady state with some oscillation that varies in amplitude based upon how closely the GPS and IMU agree.
I know on our bots when they do rapid direction changes the variance goes way up because the lag from GPS means its always predicting motion in the wrong direction but in a straight line it drops a lot both because GPS is more accurate under those conditions but also because the inertial data matches the GPS track.
Good call. I had the same thought with respect to the 0 covariance with GPS.
Right on point! As the GPS model will not me very precise and determinist it could be nice to take a look at alpha beta filter.
Interesting to see the board temperature decrease as it gains velocity.
he should add a heater and an insulating enclosure. It will decrease the drift in sensor readings and thus reduce inaccuracy both within a flight and between flights.
At 8:40, I had absolutely no idea what was coming, and I didn't even know what it was when you said it. In fact, I didn't understand most of this video, but it's still entertaining enough for me to watch the whole way through.
" we're about to cross 200'000 subscribers if we haven't already"
i look down to the subscriber mark: 200.000 on the mark
nice timing
Technician: ... um ... Flight ... we're getting a slight oscillation in the 'Z' axis.
Flight Boss: ... hu? ... Oh ... it's ok, that's just J.B.'s signature "Z axis wiggle".
Technician: (puzzled frown) you mean he designed it to do that?
Flight Boss: ... um ... ... yes ... yes he did ... now pay attention to your board.
Congratulations! I watched the entire AVA assembly to this point (many of your other videos as well).
Your persistence is contagious. It helps keep me motivated for my own projects. Hopefully, in the near future, they'll be a legitimate sustainable business.
As practical and frugal as you are, I really appreciate the effort you put into aesthetic details, from your video editing to your PCB design.
"Let's say it at the same time, I know what you're thinking, ready, ok, it's"
Me: "Another cool graph"
Joe: "The process covariance estimates"
Me: "Yea, that's exactly what I said"
Love the overlay with all the data. Not that I really know what the significance of most of them are lol
It need to be embeded in a SpaceX style overlay... simply displaying easily read figures for speed, altitude etc...
Wow thank you so much Joe that means a lot to me @Aerospacefan46 and also I’m sure Charles definitely passed out by viewing this !!
Always awesome! Very impressive work developing the AVA board, those data records are beautiful.
Fantastic flight! Thanks for not only sharing the video of the rocket, but also breaking down the data in such an easily understandable way. I showed this to my wife (1st time viewer) and she was able to follow along just fine, which is a real testament to just how well you explain things. Good job!
Congratulations on 200K! Keep up the excellent content!
Congrats on 200k Joe! Well deserved
Hi bps, just love that I don't get giddy watching Sprint onboard vision. No spin - me grin.
Well done and look forward to when you Scale UP
This data vis and video format is absolutely stunning
Yes, it’s joey again! Love the rapid launch and video schedule due to the current... *pandemic*
Planning on building my own rocket, inspired by you! The reaction control tutorials are very helpful
You have definitely been an inspiration to many people :)
I also plan to make a liquid fuelled rocket after the practice with homemade (very large) sugar motors.
:)
:)
:)
:)
:)
2 videos in 2 days!! and you've reached 200k :D Congrats!! Keep up the good work!!
Awesome! Love that your promoting other rocketry channels. It shows how much your confidence is building.
You should get that vertical shot from one more angle to triangulate Sprint’s flight angle.
I was thinking the same thing. I did that for a science fair experiment years ago to track the rockets stability
It’s just money, another camera
Congratulations on 200k! Keep up the good work!
the analysis was super interesting. I'm having a look at this plotting program
what is name of analyhze program
can’t wait to see this thing hopefully go 1 kilometer up . Love the work joe 🦀🦀🦀
It's people like you that will blaze the trail once interplanetary travel becomes wide spread. Keep up the great work!
Bravo Joe, Bravo.
Congrats on 200k man
amazing footage and analysis - love your videos! salute from germany
“joe, why is that?”
THANKS FOR ASKING!
@@BPSspace xD LOL!
Congrats on a nice flight! I really love your scrappy methods to get the job done, including the use of open source tools.
Congratulations on another successful flight! I love watching the launch with all of the data surrounding the screen!
Fantastic as usual, Joe. Really enjoyed the detailed information about the operation of the flight.
As an aside, since I'm working on some pretty over-equipped environmental sensors right now and have sensors on the brain: have you considered the use of magnetometers to aid in orientation sensing? They're available in very small footprints (e.g. NXP MAG3110, which is 2x2mm DFN10) and in my experience they offer pretty decent measurements at nearly 80Hz, which might help you improve your integrator accuracy during alignment correction, and better sense pitch during flight.
The noise isn't the best on the MAG3110, at about 0.3uT in practice (so 0.6% error on a standardised 50uT earth field). That translates to about 0.55 degrees of error on an axis if you're normalising to 90 degrees per axis, but you can get more expensive alternatives like the RM3100 that offer incredible resolution and noise floors in the order of tens of nanoteslas. I figure if you've already built the Kalman filter it shouldn't be too much work to add geomagnetic orientation values into the mix and keep the covariances low.
Anyway, enough back-seat engineering from me. Good flight and I look forward to future videos :)
Congratulations for 200k, I hope you're doing well.
Woooooooooweeeeeeeee that was one stable flight 🤭
Also massive thanks for the shoutout! and congrats on 200k :)
By now I have started to back your launches as much as spacex ones. This is more special.
Happy 200k subs :D
Good to see the normal uploading back!
Thanks prof. Joe. Amazing science, dedication and iterative development. Just a dream! Keep it up!
Well done buddy. I love the data outputs.
Nice Video and project
Congratulations for 200.000
Thanks for showing us this data!! Interesting to see all the data you get back from your flights!
That was sexy, i think i got a chub. Its been real great watching you from the start to now and all you have accomplished! Keep it up!
200K!!!! Congrats!! 🎊🍾🎉🎈
Awesome! Super excited about the progress!
Wow you just hit 200k subs, congrats!
Excellent, as always.....
Still working on first vehicle, of my long delayed return to rocketry. Nothing fancy.
Beautiful, once again proved: Mistakes are no mistakes
There are no mistakes, just happy accidents.
"May your skies be blue..." gets me every time. Same to you!
Ok this is waaaaaay more satisfying than it should be.
_"I love it when a plan comes together...."_
😊😊😊
loved the tracking animation, really worked.
So, we agree: Nothing happens to the board during welding!
Greetings from Munich, Tom.
Do I spy an RC plane-launched model of Virgin's launcherone in the background?
Please Joe. Make that happen.
Congrats on 200K Joey!
BEEPIS SPACE.
I like that you take this so seriously. The debriefs are awesome
One of the things that I like about this is someone could find this some time in the future and it will still be useful not just a cool launch but the data to go with it. *So cool!!!*
no one does anything alone. good flying, dude!
I love the people cheering in the background. They sound so excited.
On the topic of servos, the RC helicopter world has several reasonably priced strong lightweight high speed brushless servo manufacturers. I specifically bring them up because the servos on the heli swash plate have to have a consistent response speed. or the models become uncontrollable. BK servos in particular are spectacular.
Pretty amazing stuff. Glad to see your system finally working. I know it's been a frustrating journey at times.
Love the videos! I am learning from every video of yours that I watch and I am trying to build one myself!
Keep it up!
Amazing work! I love seeing stuff like this it really sparks my curiosity!
Really really appreciate it
This is so prefect and straight if you have seen his works before this，That is amazing！！
Dude, i didn't understand a word you said. However, the video was AWESOME! Great job!
Hey Joe I am from India. Insert A.V.A. into falcon heavy. And your rocket data is so good. Congratulations for success....
5:03 What is that "Raw LOL (?)" You are mentioning?
He's talking about Raw LLA, which stands for Latitude, Longitude, and Altitude. It would tell us exactly where the launch field is.
THAT makes sense :D thanks
Now it needs to land itself. Great job 👏
Love the "xenon" lights you have set up :)
Congratulations on 200k subscribing
This community is great
Great work and fun!
Fantastic.
You know... Joey B is a class act!
Sir,great video as always,A AM A NOOB AT THIS STUFF,but here is an idea, maybe use a high res quarature rotary encoder to characterize the servo(or make a lookup table) and use the pot of the servo with an external high speed ADC with a different micro controller for total control over that TVC. Man That KF looks so much better. Best of luck !!! TH-cam lacks content of this caliber and quality and fun and SO MUCH FUN WOW.
Hi Joe,
Have you thought about using your camera data to map the flight path in a 3d program?
I work in CGI so can point you in the right direction and help trouble shoot any problems you may have if you want to go down that route.
Cant wait for launch 9
*Great. I Wonder How Many Mathematic Awards Did This Guy Achieve*
I came from the Motherboard video. Damn I’m hooked now!
Great feeling when it works
Would you ever use a hybrid rocket motor? (Not sure if you’ve answered this one) also congrats on 200k subs 👍
Have you ever considered trying something like the zenit flyback booster? Or is it too complicated to teach the computer two different flight modes?
I can't wait flight 9 and 10.
Congrats on 200k!
Well done keep up the good work 👍👍👍
This man is essentially space X now.
Oh yea! Nice job man!
Great work Joe !
cant wait to see you build the starship model.
One of your videos was shown in a TV program in my country. The rocket fail one!
Incredible stuff, I love your channel. Would it be possible to determine the wind velocity versus altitude profile during the first half of the flight as you fly? Could that information then be used to target a trajectory that would have the parachutes drift you back to the launch pad?
I like the variable naming convention.
Yay back to back videos
That was a beautiful flight!!!
Congrats on a successful launch!
Looking forward to the longer flights, should be awesome
3:50 “flight we’re getting a little shimmy”
I was looking forward to this one
dude which provide more stability tvc or fin control like in missle in model rocket. can u make more projects on fin contact as it would be better for people since they have cheap rockets and servo need not take that much stress
Great to see a successful Sprint 8 flight. Looking forward to Sprint 8 and 9. Achieving 1Km. altitude will be amazing. What is your goal after the 1Km flight?
Does someone have a link to "Curve"? I've searched and can't find anything actually called "Curve".
It’s getCurve.io
@@CCSTechno Danke
@@CCSTechno Thanks!
I love Your work!
Bravo Elon Joe.....I am learning all the videos of yours that i watch. Greetings from E.Pakistan 🇵🇰🇺🇸🇵🇰🇺🇸🇵🇰🇺🇸
Congrats! Can't wait for the next one!