136 points by scottshambaugh 2 days ago | 161 comments
krisoft 2 days ago
Mixing up control errors with absolute errors is a very common form of miscommunication in robotics.
I work with relatively big robots and often my colleagues would say something like this "During the test we had 0.5m cross track error, so we did X, Y, Z ...".
And I always ask them for clarification. Were they looking at the robot and seeing that it is half a meter off where it should be, or were they looking at a screen and seeing that the robot thinks it is half a meter off from where it wants to be? Because those are two very different situations. And both can be described with the same words. (And sometimes it can be both, or just one of them.)
meindnoch 2 days ago
mucle6 2 days ago
killjoywashere 2 days ago
yetihehe 1 day ago
rcxdude 2 days ago
It's definitely floated around for a while, but it grew in popularity in the past few years.
lacrosse_tannin 2 days ago
hedora 2 days ago
ls612 1 day ago
scottshambaugh 2 days ago
This always stuck out in an otherwise excellent bit, because you should definitely _not_ be taking the absolute value of your control error.
m4rtink 2 days ago
rascul 2 days ago
amelius 2 days ago
Control error is defined as the difference between desired value and measured value. So this is pretty good?
Even if they use some crude method to obtain position (e.g. gps), they can still easily refine that using e.g. triangulation using cameras around the landing platform.
krisoft 2 days ago
Not sure what you are talking about. If you are asking if 0.5cm is good controller error for an orbital class launcher on landing? Yes, it is extremely good. Without doubt.
If you are asking about my tangential story where there is confusion between total error vs controller error then no, it is not good. Confusion is never good. Especially if the system is not within the total error budget. Because to improve it you need to know if you are dealing with measurement error or controller error.
> Even if they use some crude method to obtain position (e.g. gps), they can still easily refine that using e.g. triangulation using cameras around the landing platform.
Sure. I doubt that their total error is within 0.5cm, but both of their landings were extremely succesfull.
thot_experiment 2 days ago
I assume they are doing something much more clever/hardened, but you can trivially achieve much greater spatial accuracy with a Vive Tracking Puck for like $100.
MadnessASAP 2 days ago
Your positioning system needs to acquire a fix at least 100m out in variable atmospheric conditions on a rocket undergoing heavy acceleration and dumping all kinds of heat, smoke and vibrations into itself, the atmosphere, and everything around it.
In addition having a fix on your tracking device is only half the game, not you have to figure out where the rest of the rocket is in relation to your tracking device. Which again, vibrations, temperature and manufacturing all have an effect.
So while yet, a vive tracking puck isn't entirely unlike the workable solution it is also entirely unsuitable as a solution and should not be used as a baseline to measure off of.
thot_experiment 21 hours ago
cubefox 2 days ago
asadotzler 2 days ago
cubefox 2 days ago
Sammi 1 day ago
Looking at the image I can see a dark device on top of a mast that could be anything electronic.
cubefox 1 day ago
jameshart 22 hours ago
sebzim4500 2 days ago
i.e. the booster doesn't know it's actual position to within 0.5cm but it knows it's position relative to a buoy or the catch arms to that precision.
beerandt 2 days ago
But survey grade gnss is a web of rabbit holes, if you want to get into it.
And there are ways to get sub mm accuracy both relative and absolute, but idk of one that would be quick enough for the required reaction time of dynamic landing via 'catching'.
But multi-centimeter (4-5) that's really easily doable is probably good enough for other systems to take over from.
magicalhippo 2 days ago
[1]: https://www.youtube.com/watch?v=pAPt5vbr-YU (don't recall timestamp, sorry)
bmicraft 2 days ago
snypher 2 days ago
bmicraft 1 day ago
GNSS is more than accurate enough once you know all slight errors in satellite orbits and the atmospheric distortions currently affecting the area near the base station and can correct for them.
cubefox 2 days ago
rkagerer 2 days ago
You can see the arm comes in, then there's some side-to-side bounce (not sure how much is the rocket bouncing off vs. the arm fine-tuning its position). Just after contact seems to be made, and before the shock absorbing (or yaw-correcting) pistons drop much, there's a large flash from the engine. Is that a characteristic of engine shutoff, or was there a last-second "hover" push just before shutoff and drop? I wonder how much force the arms felt.
Another perspective showing both arms, and (as mentioned in the article) how the left one adjusted more significantly at first: https://youtu.be/JlcrNakUGVs?t=3
HarHarVeryFunny 2 days ago
https://nextbigfuture.s3.amazonaws.com/uploads/2023/12/Scree...