Why Japan's Moon Lander Crashed Due to An Unbelievable Computer Bug

foreign Scott Manley here late last year SpaceX launched the ispace hakuto r lunar lander on a Falcon 9 rocket it would take a few months to get they are performing a low energy transfer via the Earth's sun LaGrange points once in lunar orbit it spent some time checking systems out before finally beginning its descent towards the lunar surface with the intent of making a soft Landing so it could deploy its science instrumentation and other cargo as you probably know by now this is not what ultimately happened during the live stream of the landing we saw the Telemetry showing the spacecraft slowing itself down as it went lower and lower until it was apparently just above the surface and showing a speed of only a few kilometers per hour perfect for that beautiful slow touchdown on the lunar surface which was surely just right there now at this point we just got a simulation uh and ultimately we found out that the Lander was lost and in the last month or so there's been a fair amount of speculation and information has come to life first of all there were the radio amateurs who watched the spacecraft's radio signal and they noticed that after a certain point it began Free Falling before the signal was lost this suggested that the spacecraft had stopped moving some altitude above the surface before then falling to the surface and impacting at a speed of about 500 kilometers per hour the lunar reconnaissance Arbiter was able to image the site just recently and could compare the images of today versus some time ago and showed a number of well changes in brightness which are presumably effects of debris you know hitting the surface and scattering there isn't a new crater generated by this because the impact velocity would have been low enough that the vehicle tended to break rather than make a hole in the surface unlike say the Landers from India or Israel so we have a pretty good idea of how the spacecraft ended up hitting the moon fast we're just not sure why this happened until this morning when the team published a report which pinned down the blame on a software issue specifically the way the spacecraft would fuse data from multiple sensors to determine its location in space this is a core problem in space flight when we play a Kerbal Space Program and we're performing lunar Landings we have instruments on the screen which are actually perfectly 100 accurate but in real life a spacecraft has to take information from multiple sensors it has accelerometers and gyroscopes which measure the rate of change of acceleration of a velocity or the rotation it may have radar altimeters it might have radio information from the ground which is giving a Doppler information now you have to kind of synthesize this into the the other state Vector into the position and velocity and one of the interesting things about this problem is that things like accelerometers they tell you how fast your speed is changing whereas radar might tell you how high you are and perhaps how fast you're going in the vertical Direction so you have to try to take these disparate ideas and turn them into one unified version of reality according to the laws of physics so in general you call this Sensor Fusion one particular method that's used is called Kalman filtering and on top of this this is an autonomous spacecraft that needed to operate in real time without human intervention and it needed to be able to deal with potential Hardware failures and then continue to try you know try to complete the mission to the best of its ability so according to the team what happened was during the approach towards the landing site the vehicle passed over a crater with steep sights and all the time the radar altimeter is paying this thing and it sees the altitude suddenly change by three kilometers as it passes over a steep cliff and the software says wait a second the rate of change of the altitude shouldn't be this much I think that this SEC the sensor is broken and so like any well-programmed piece of software it began to ignore the spurious data except that the data was actually correct now without the radar altimeter it no longer had a good measurement of where the surface is while it could estimate it based upon its knowledge of the the shape of the Moon its accelerometer information the you know the Doppler data that it was getting from the earth it didn't have that one piece of data that specifically told it where it was and so it was measuring distance to the surface by basically understanding that it was initially moving at a certain speed in a certain altitude and it was slowly subtracting off that altitude by measuring results from the accelerometers and Gyros so the result starts to drift away from the actual number and on top of that it will very likely carried in like an error just because of the time that the uh the altitude was sampled they were headed into a crater by the way so they had to fly over this crater wall and then descend into the crater which was lower than the surrounding Moon here's a map showing the sort of rough context uh the spacecraft was coming in from the north and it would have flown over that Ridge and then had to go downwards so knowing the altitude that it was hard targeting to was kind of important regardless the spacecraft thought the surface of the Moon was about five kilometers higher than it actually was and so the Control software did a perfect job of setting this thing down at its altitude where it thought the lunar surface was and then it just sat there balanced on its rocket mode or slowly descending waiting for the moon surface to come up and greet it and before long it ran out of propellant at that point it actually ran out of repellent asymmetrically and it began to spin by the looks of things but yeah it basically fell those five kilometers to the surface so the next question is why wasn't this Behavior caught during testing and stopped they could fly simulations over the surface with you know with a you know synthesized inputs and verify that the software was behaving well uh they did do these simulations and they did it for various Landing sites that they had planned but at some point the planned Landing site changed and the simulations had all been done and were completed by that point so they never caught the fact that flying over this particular Cliff would ruin their day so I can imagine the next time they fly they're gonna have performed a proper end-to-end simulation of all of this and probably fix this particular bug in the software and you know like this is actually a thing that we've seen before right these kind of software bugs which lead to the spacecraft not knowing where it is uh happened a previous time that I remember is the skiapareli Lander by from the Europe and what happened to it was that during in Parachute deployment it bumped around and because the rotation rates in the spacecraft exceeded the rotation rates or allowed it ignored them and then when it came back and read the radar the spacecraft thought it was upside down and because the radar said it was like you know 12 kilometers up or whatever it then multiplied that by its facing vector and it said actually I'm underground so I should cut my parachute and so it then cut the parachute and fell to the surface rather like this this is the third time we've seen a small team try to put a Lander on the surface of the Moon and fail and every time it has been due to Software India's Vikram Lander went out of control a few hundred meters above the surface due to software uh Israel's bearishit Lander had a software glitch which caused a computer reset and during that reset it wasn't firing its engines so it ended up plowing into the moon going too fast at this point I'm really hoping the astrobotic has gone over its software with a fine-tooth comb and a lot of simulation because it would be nice to see a small Lander landing on the moon softly for once I'm Scott Manley fly safe [Music] foreign [Music] foreign [Music]

Channel: Scott Manley

Views: 902,837

Keywords:

Id: 2JlUnOAiMm4

Length: 8min 38sec (518 seconds)

Published: Fri May 26 2023