Hacker News new | ask | show | jobs
by yoursunny 1805 days ago
It's amazing how much redundancy was built into those out-of-touch systems, 30 years ago. However, the number of backup units is finite, so let's hope the now-operating system can last for several more decades.
3 comments

The part that caught my eye was redundancy on memory.

>> Hubble’s operators initially thought a memory module was at fault but switching to one of three backup modules produced the same error.

Apparently Hubble has 4 memory modules which are switchable! I'd love to see how that works. Actually, I'd be fascinated to get a walkthrough of the overall architecture. It might give insights for how we keep business continuity by first accepting that hardware and software will fail.

Posted this before here. Maybe worthwhile to post again...

Fig 5-10 is the Data Management Subsystem

https://asd.gsfc.nasa.gov/archive/sm3a/downloads/sm3a_media_...

Concerning the computers:

- First they had a DF-224 flight computer and a

- Science Instrument Control and Data Handling (SI C&DH)

Initially DF-224 between missions got installed a coprocessor:

https://asd.gsfc.nasa.gov/archive/hubble/a_pdf/news/facts/Co...

During another servicing mission they replaced it with something called the Advanced Computer with Intel 80486:

https://asd.gsfc.nasa.gov/archive/hubble/a_pdf/news/facts/FS...

There are some 50,000 lines of code in the C and Assembly programming languages. https://www.nasa.gov/pdf/327688main_09_SM4_Media_Guide_rev1....

They also have a Help Desk...

"Welcome to the Hubble Space Telescope Help Desk"

https://stsci.service-now.com/hst?id=hst_index

Classic movie scenario. Hubble's down, and we need it now. No-one can solve the errors. Old guy walks in, worked on Hubble 30 years ago.

"There's a backup module, with an override command to activate it, but it won't work with the system down. You'll have to use the manual override switch - on the telescope."

It was 64k of core memory originally, but it was later replaced by redundant 4x 64k CMOS memory.

https://youtu.be/RWUnC2uf3XY?t=406

…or as I try to occasionally discuss here… let’s simply develop the unmanned robotic capability to service Hubble, etc

I found one discussion from 7 years ago

“… I think we should start a more extensive national unmanned space program. For example, if the Hubble, or its replacement, needs to be fixed, we should have an unmanned answer, for instance.”

https://news.ycombinator.com/item?id=8540664

Ah yes, let's simply do something very hard.

Huble was never meant for this, if you look at all the systems present on ISS for automated docking, none of that is on Hubble. Access to modules inside Hubble was never meant for robots with very limited dexterity either. Being able to do something like that would be a huge feat of engineering in my opinion (and extremely expensive).

I am fairly certain it would be faster and cheaper to just build a new one from scratch.

And even if you managed to make a robot to service Hubble, it then would only be able to service Hubble and nothing else. JWT for example is completely different.

On the long term you are right that this is a capability we need, but this needs to be taken into consideration while building the telescope/satellite/whatever: automated docking mechanism, standard ports and dimension of parts etc. etc.

I'm pretty sure the pitch here is "a robotic being with human perception, dexterity, and manipulation, but who doesn't breathe air and never gets tired."

So the idea is that you don't need to specialize it to the thing it's meant to work on, because it works on whatever a human works on. A similar idea drove a lot of the DARPA Robotics challenge, with its emphasis on being able to drive a normalish vehicle, open a door, climb a ladder, use a regular power tool, etc.

Anyway, I think the state of the art for all this is still pretty far away, which is why the instinct is to assume we're talking about something specialized.

> I think the state of the art for all this is still pretty far away, which is why the instinct is to assume we're talking about something specialized.

Yes, that's exactly what I thought. If we are talking about AGI + human level robotic dexterity, then the use of "simply" becomes even funnier.

I'm not sure about AGI. You can just control the robot remotely from earth. Sure, there will be a minor lag but definitely won't justify need of AGI.
NASA was experimenting with this in the past with Robonaut (apparently it also started as a collaboration with DARPA)

https://en.wikipedia.org/wiki/Robonaut

I believe Robotnaut is back on earth as of sometime in 2018, but it was briefly at the ISS and even ran ROS— there was an official NASA-supplied Gazebo simulator for it: https://www.slashgear.com/nasa-robonaut-2-simulator-stack-no...
> Access to modules inside Hubble was never meant for robots with very limited dexterity either.

It was also never meant for humans with less limited dexterity.

I recall one of the Hubble servicing missions I watched on NASA TV, in which they had to bolt a special adapter plate over a cover, unscrew over a hundred tiny non-captive screws (which the adapter plate was designed to catch, so they wouldn't float away), and only then could they open that cover. That part of the telescope clearly wasn't designed to be serviced in space.

To be fair, they wasn't expecting Hubble to last beyond the end of the service date. I recalled it was ~10~ 15 years? They have no way of knowing this beforehand.

Edit:saw the comment somewhere that it is 15 years.

I think 'simply' is precisely the wrong adjective.

Plus you'd still need spare parts. Why would I build a robot to swap parts, when I can simply put all of the spare parts into the telescope and swap over to them electronically?

<pedantic> adverb </pedantic>

The other thing to remember was that Hubble was designed in an era that the Space Shuttle was meant to make manned missions to repair / upgrade commonplace. My understanding is that this was actually done at least once to Hubble (maybe more? I forget); but unfortunately for Reasons, NASA turned out to be incapable of resisting cutting corners which put people's lives at risk (Challenger, Colombia). A system designed today would be designed assuming that it would be robots or nothing.

I was curious myself, and it turns out there were five missions, starting with the one to compensate for the incorrectly ground mirror. IIRC the final one was in doubt, as there was no possibility of a rescue mission in the event of irreparable damage on launch.

https://www.nasa.gov/mission_pages/hubble/servicing/index.ht...

The final servicing mission (STS-125) ended up with a contingency plan for damage on launch, STS-400. Another shuttle was prepped and on the pad for a rescue mission if the primary shuttle was damaged.
Knowing what will go wrong is hard to predict. Having the capacity to have robots service satellites, etc would be extremely useful.

Also, the vision is that robots can build on Mars, for example.

“Simply” means fund the research. My comment was from 7 years ago.

We also need the robots here on earth for dangerous tasks

We’d make impressive progress over each decade with more effort.

> Why would I build a robot to swap parts, when I can simply put all of the spare parts into the telescope and swap over to them electronically?

What if the system that swaps parts fails?

What if the robot fails?
> What if the robot fails?

Touche. At least you can send another robot!

But in reality, the ideal system is probably a combination of built-in redundancy / spares for parts that can work that way; and then robots for the kinds of things that are too complicated to have a built-in failover.

The robot fails to perform the task or fails reaching orbit,etc?

Send up another [improved] robot. These missions are less costly than human missions. We get to iterate more often

> Send up another [improved] robot. These missions are less costly than human missions.

Says who? And how? This seems like an unsubstantiated claim.

We can put much larger payloads into orbit now and will be even moreso with Starship.

It would make sense to start working on a replacement for Hubble, even if that means it'll be ready in 20 years.

Or we could build a cheaper telescope that is less redundant more quickly, and launch a new one every 5 years. Launch prices are falling, so this could be more economical than building expensive long-lasting telescopes.
There's several problems with that approach. The first is there's going to be a minimum size for a telescope to be useful as a scientific instrument. For a telescope intending to be in any way a contender for Hubble's mission it would need a primary mirror a meter or more in diameter. Then it needs reaction wheels for precision aiming, RCS, instruments, electronics, and power.

At a minimum you're outside of "smallsat" size range. Even building a lot of them the precision manufacturing of the optics and steering systems require a lot of ground based testing, calibration, and qualification. Between the size and precision even your "cheap" Hubble's aren't all that cheap.

The second problem is terrestrial telescopes have gotten really good and are far cheaper for any given mirror size. Their instrumentation doesn't have nearly the same mass restrictions and can be swapped out routinely.

Space telescopes make sense when their mission requires them to be space based like observing wavelengths that don't reach the surface of Earth.

Also swarms of telescopes that can collectively take higher resolution, broad spectrum photos.
While we don't have a direct capability to do this now, some of the large programs of record are being built so that robotic servicing may be possible.

JWST has a docking port for a future robotic servicing mission.

Nancy Grace Roman (formerly WFIRST) space telescope has grappling points on the spacecraft for robotic servicing.

Pretty sure it's much easier and faster to do it with humans and Starship, assuming the necessary interface is developed.
starship wouldn't really fit the mission profile of sending 2-3 astronauts to hubble orbit for EVA. The same starship could instead deliver a fleet of replica hubbles to lagrange points autonomously.
Or built the telescope into Starship and use it as an observatory.
I agree in a more general sense: space is a natural domain for robots, not humans, and we should be designing out space flight capability with that in mind. We have some almost 50 year old robots advising us on the heliopause right now.

The general problem of repairing Hubble robotically is unlikely to be solvable within the lifetime of the telescope, but future devices could be designed with that in mind.

Why don't we just build 5 Hubble copies and send them up? Should be much cheaper now.

Is the answer is that NASA doesn't know how to build them anymore? Or that it's not politically feasible?

Why don't we just build 5 Hubble copies and send them up? Should be much cheaper now.

"Just" is being extensively overused in this thread.

If all of them suffer from the same issue (let's say memory corruption), you end up with 5 satellites that will fail on the same timeline.

When operating multiple devices to improve reliability, diversity is one point.

NASA doesn't know how to do anything cheaply. They have two space telescopes sitting in a warehouse since 2011 that were donated by the NRO. They're estimating $4 billion to launch and run just one of them, more than a decade after the donation. And you know it'll cost double the estimate and take an extra 5 years.
Why dont they just give others to do their work? Last I remember india sent a probe to Mars for less than amount it took for making the movie Mars featuring Matt Damon ?
For the politicians that fund NASA, spending money is the whole point. They call it "job creation".
agreed but what job creation is a project sitting dead for 5 years? are they giving people salary for not doing anything and then claiming too much costs or something else?
Considering Hubble was an old NRO design that they donated, I bet there are more backup parts than are publically known to be available.
Just to clarify: Hubble was not an NRO donation, although it bears some similarity to the KH-11, an NRO satellite from the 70s. It is not confirmed whether the two actually share parts, as the KH-11 is still classified.

The NRO did donate two unlaunched telescopes to NASA in 2012 (with optics present but sans electronics), which still have not yet been retrofitted and launched.

If you read up on it, the NRO gave a shell for a Keyhole something to do NASA. It had no optics and no sensor AFAIK, that was on NASA to build.