SpaceX has taken an interesting approach here -- effectively, they've crowdsourced a more-robust MPEG decoder! It would be interesting to hear a description of the steps they've already taken to repair the bitstream (and would be really cool to see source for it). I fear that may be asking too much, though.
I also wonder how much could be recovered if they had a sufficiently advanced soft-decision air interface. If the missing macroblocks from the bitstream are simply macroblocks that have failed a checksum, data from the radio receiver as to which bits it has the lowest confidence in could be very valuable in pulling more blocks "from thin air". The answer may be "no, there is nothing else to recover", but it seems like a good jumping-off point for more research...
Looks like it's a good few thousand feet in the air still based on those white caps. They look like what you see from a plane. But it's pointing in the right direction. Very cool.
In the raw video, the engine shuts down at 0:23, and before that you can see the engine blasting the water. It's not coming down that fast, so I'm pretty sure it's a lot lower than a few thousand feet when that good frame appears earlier.
This had to be so disappointing when they pulled the footage out and found it.
You can kind of vaguely get a sense that it worked from the picture though: the motion in the compression artifacting shows the engine burning till landing, and then you get the distinct "tipping over" sense afterwards before it all goes grey.
Bummer the quality is so poor. I was really hoping for something really, really cool to watch. I'm guessing this is more than just a GoPro taped to the side so it is surprising to me that the video failed so hard. It also surprises me that they didn't have several cameras just in case. But in the interest of staying positive, I guess they learned a few more things to avoid next time. I am super excited about the progress SpaceX has made and will continue to make. I can't even count the number of times I've watched the "low launch and ground landing" videos. So cool to watch history in the making.
Presumably the real interesting data comes from the sensors, and the camera was just an add on, possibly only for publicity with no real technical value, so they did not go to great lengths to make sure it would record well.
When you have a large dynamic system you always have different kinds of sensors. You have modeled sensors that tie in directly to the feedback loop controlling the device, and you have "wtf happened ?" sensors.
There are 2 big considerations:
1) simplicity of the sensors. Meaning how easy is it to use the output of the sensors to get to a useful model of the situation. Accelerometers, what you'd call "sensors" are somehwat easy (though nowhere near as easy as you'd suspect, read up on kalman filtering for a few of the considerations)
2) comprehensiveness of the sensor information. Meaning if you've got just the output of this sensor, how much do you know (after careful analysis). Accelerometers only give you one aspect, so they fail pretty badly.
Cameras are definitely comprehensive sensors. They are really good to understand afterwards exactly what happened. Not so easy to write control loops against.
The fun thing is that nature has diametrically chosen for comprehensive sensors, coupled to a nervous system that can learn extremely complex feedback loops. Feedback loops so complex that you'd have zero hope of ever modeling correctly in a lifetime. Humans move based on information from one eye (not even 2 eyes, but sometimes the left hemisphere controls the body based on information of the right eye, sometimes the right hemisphere controls based on information of the left eye).
Now if you look at the control tasks that nature can carry out, versus what we can do using technologically, we are pretty pathetic. Bats, for example, fly at 50-60km/h right through the leaves of bushes and trees in pursuit of insects in total darkness (using sonar) without impacting any of the branches, they locate insects on the other sides of branches or leaves without even looking, and they do this with thousands of interfering signals sent out by independant mobile platforms [1]. They can fly through a bush as if it's not even there.
The limit of human technology when it comes to sonar are ultrasonic distance sensors [2] ... there's no comparison.
So yes, technology uses sensors, but we really shouldn't. It's not getting us anywhere, at least not in mobile robotics, like rockets. Trouble is, it's such a very very hard problem.
The problem wasn't the camera, it was receiving the telemetry. As Elon said after the launch, they captured telemetry from a plane, and the plane had to stay pretty far away.
I was wondering why it was taking them so long to post a video. Either post it or don't, but a delay is just weird.
Well, now it all makes sense. Terrible corruption! Unfortunate, but these things happen. Certainly a minor failure compared to the overall great success of this test.
I also wonder how much could be recovered if they had a sufficiently advanced soft-decision air interface. If the missing macroblocks from the bitstream are simply macroblocks that have failed a checksum, data from the radio receiver as to which bits it has the lowest confidence in could be very valuable in pulling more blocks "from thin air". The answer may be "no, there is nothing else to recover", but it seems like a good jumping-off point for more research...