4 ms·

Hey, I'm glad you responded, and that I misinterpreted where you were coming from and your intentions, but I guess that happens when limited to text (not that a

by natep 14y ago

Hey, I'm glad you responded, and that I misinterpreted where you were coming from and your intentions, but I guess that happens when limited to text (not that a misunderstanding-free medium exists). But as you've probably gathered by now, most of my response came from reading "I don't see any reason why someone shouldn't make this. [and I doubt a good one exists]" instead of "I don't see any reason why someone shouldn't make this. [and I would like to hear some]" > Fast, Cheap, and Out of Control The first thing that popped into my head when I saw this title was NASA's "Faster, Better, Cheaper" initiative. I will admit that I do not know much about it, but what I do know is that there were some failures (as you would expect) and the public did not react well. The failures did not come from not having space-grade parts, IIRC, but for various other reasons. The most infamous was the Mars Climate Orbiter, known for failing due to an imperial-metric conversion error [1], that is still brought up in almost every space-exploration piece that gets sufficient attention. Anecdotally, the lessons learned by those I've talked to are (1) choose 2 and (2) the public will not tolerate failure on large NASA projects, even if those projects cost a fraction of what it takes to host an Olympics, or to buy Instagram, or fight a war for a day.[2] But I have just now hopefully started a discussion with my coworkers about this paper[3] on our message board. So anyways, about the paper. It looks like your idea is only mentioned briefly in the one section, and not fleshed out. The idea was somewhat more fleshed out in the talk I went to. What I meant by my line of questioning was that I haven't seen an implementation of these concepts outside academia, even though the idea has been around for some time, and the industry isn't entirely driven by irrational beings, so there must be some technical reasons why the ideas haven't been fully adopted. >> * Having an identical copy is not redundancy > Is it because the failure points remain the same? Essentially, yes. His argument that most failures were systematic, and not due to unequal 'wear' of various kinds. Software especially, since on space missions it is implemented to be as deterministic as possible. That means that even if the primary processor fails gracefully due to bad internal logic, and a hot backup immediately takes its place, the backup will behave in exactly the same way given those inputs, which are likely to stay the roughly the same across the switch. Another example is a mission where the high gain antenna succumbed to a systematic failure, and they completed the mission on its low gain antenna. If there had been 2 identical antennas, they would not have been able to. Designing and testing components to last a long time under the conditions you expect and test for is relatively easy. It's when the designs and tests don't match reality is when the problems happen. If you write something and make a copy, it will retain all of the typose of the original, and it's the same with Code/CAD/etc (and you might be more likely to introduce bugs than fix them). Ground-based systems can get away with this because of how easy it is to replace/repair broken units in a redundant setup. [1] Even though it was due to a much larger issue related to how the project was carried out, and just happened to manifest in that error. It could have easily been a kilometer/meter mixup instead. And having a backup string of landing hardware/software wouldn't have helped in this case, unless the backup units had a different design/implementation. [2] And yet, JPL still chose to use the skycrane. Very couragous. [3] http://www.dau.mil/pubscats/ATL%20Docs/Mar-Apr10/ward_mar-apr10.pdf http://www.dau.mil/pubscats/ATL%20Docs/Mar-Apr10/ward_mar-ap...

todayiamme 14y ago
>>> most of my response came from reading "I don't see any reason why someone shouldn't make this. [and I doubt a good one exists]" instead of "I don't see any reason why someone shouldn't make this. [and I would like to hear some]"<<< Ah I see. I'll start qualifying my I-don't-see-any-reason-whys from now on with the latter statement. :) >>> I will admit that I do not know much about it, but what I do know is that there were some failures (as you would expect) and the public did not react well.<<< I guess that's the real reason why this won't be implemented by government agencies. Space missions are a matter of national pride and no one wants a "designed to fail/waste of money" accusation on their hands. However, empirically speaking it seems to be the best way to do things as we can increase the amount of explored area quite rapidly and respond to changes much more quickly. I think that when private companies such as Planetary Resources start doing exploration they will be forced to adopt this model because of their constraints and a lot of amazing solutions will come out of it. If that happens then it might open up a pandora's box and advances will happen much more quickly. Some people will see it as a bad thing, but as long as the systems are autonomous it should lead to a lot of good things. >>> 2) the public will not tolerate failure on large NASA projects, even if those projects cost a fraction of what it takes to host an Olympics, or to buy Instagram, or fight a war for a day.<<< More importantly the politician who approves it probably won't get re-elected. >>> [2] And yet, JPL still chose to use the skycrane. Very courageous.<<< I was quite shocked when I heard it worked. I was willing to bet on the side of failure because of the sheer complexity involved. A small timing error, sensor glitch or the other million things that could have gone wrong would have led to failure. It's quite impressive that they managed to do such a high stakes real-time task more or less autonomously. It really was quite a daring thing. >>>your idea<<< It's definitely not mine. I bet people were talking about it when I was in diapers. >>> What I meant by my line of questioning was that I haven't seen an implementation of these concepts outside academia, even though the idea has been around for some time, and the industry isn't entirely driven by irrational beings, so there must be some technical reasons why the ideas haven't been fully adopted.<<< Yes, I'm willing to bet that this has a lethal flaw which quickly led to such implementations to be rejected, but the question is can this be hacked, for the lack of a better word? Remember most organisations where rovers are designed are meant to be risk averse and they, aside from NASA, have an endless pool of resources to draw from. (I'm talking about the military) The sociological and resource incentives in place simply work against any such proposal independent of engineering viability. >>> His argument that most failures were systematic, and not due to unequal 'wear' of various kinds. <<< There's an interesting case to be made over here, one is redundancy at the unit level and another is redundancy at the systemic level. Let's take the swarm as a system, in that case if you have multiple backup copies of machines adept at performing particular tasks then you essentially have redundancy provided that the same method is not followed at the unit level. In that case if one failed for a particular set of inputs (hardware or software) then you can "patch" the rest by either avoiding the physical situation or changing the software. You can repeat that with individual units and let them have free reign. If one of them gets destroyed then the other ought to be able to be modified in time. Although it won't guard against stupidity at the unit level, this should to be a much more redundant system than anything we can create within a single one-use device. At the level of the single unit, systematic failures come into play and more copies is actually less over there. Things must be asymmetrically designed to overcome edge cases and systematic failure modes and robust engineering makes sense over there. I think that one of the best ways to implement such a system would be to spend most of the resources in creating and testing a mobile base which is independent of its payload. (attachable modules which perform specific tasks) You should then be able to deploy this machine across missions and learn from all of the real world testing in each mission to create something truly robust and reliable. Once you achieve that you can start offering redundancy at the swarm level through the payloads. For example in that high gain, low gain antenna scenario, wouldn't it be better to have a dozen robots equipped with a variety of antennas dedicated to communication? >>> It's when the designs and tests don't match reality is when the problems happen. <<< Yes, but isn't the entire point of the exercise to fail early and fail often so that you can succeed? If your system is disposable then any systemic failure becomes yet another data point to engineer against and all future systems are better because of it. There is no better laboratory than nature and surely this is a point in favour of it? (unless I'm missing something) Would you like to carry out this convo via email? If so then please feel free to drop me a line at searchingforabsolution [at] hush.com >>> [3] http://www.dau.mil/pubscats/ATL%20Docs/Mar-Apr10/ward_mar-ap.. http://www.dau.mil/pubscats/ATL%20Docs/Mar-Apr10/ward_mar-ap.... <<< Thanks for linking me to this article! It was great.

by natep 14y ago

todayiamme 14y ago

>>> most of my response came from reading "I don't see any reason why someone shouldn't make this. [and I doubt a good one exists]" instead of "I don't see any reason why someone shouldn't make this. [and I would like to hear some]"<<< Ah I see. I'll start qualifying my I-don't-see-any-reason-whys from now on with the latter statement. :) >>> I will admit that I do not know much about it, but what I do know is that there were some failures (as you would expect) and the public did not react well.<<< I guess that's the real reason why this won't be implemented by government agencies. Space missions are a matter of national pride and no one wants a "designed to fail/waste of money" accusation on their hands. However, empirically speaking it seems to be the best way to do things as we can increase the amount of explored area quite rapidly and respond to changes much more quickly. I think that when private companies such as Planetary Resources start doing exploration they will be forced to adopt this model because of their constraints and a lot of amazing solutions will come out of it. If that happens then it might open up a pandora's box and advances will happen much more quickly. Some people will see it as a bad thing, but as long as the systems are autonomous it should lead to a lot of good things. >>> 2) the public will not tolerate failure on large NASA projects, even if those projects cost a fraction of what it takes to host an Olympics, or to buy Instagram, or fight a war for a day.<<< More importantly the politician who approves it probably won't get re-elected. >>> [2] And yet, JPL still chose to use the skycrane. Very courageous.<<< I was quite shocked when I heard it worked. I was willing to bet on the side of failure because of the sheer complexity involved. A small timing error, sensor glitch or the other million things that could have gone wrong would have led to failure. It's quite impressive that they managed to do such a high stakes real-time task more or less autonomously. It really was quite a daring thing. >>>your idea<<< It's definitely not mine. I bet people were talking about it when I was in diapers. >>> What I meant by my line of questioning was that I haven't seen an implementation of these concepts outside academia, even though the idea has been around for some time, and the industry isn't entirely driven by irrational beings, so there must be some technical reasons why the ideas haven't been fully adopted.<<< Yes, I'm willing to bet that this has a lethal flaw which quickly led to such implementations to be rejected, but the question is can this be hacked, for the lack of a better word? Remember most organisations where rovers are designed are meant to be risk averse and they, aside from NASA, have an endless pool of resources to draw from. (I'm talking about the military) The sociological and resource incentives in place simply work against any such proposal independent of engineering viability. >>> His argument that most failures were systematic, and not due to unequal 'wear' of various kinds. <<< There's an interesting case to be made over here, one is redundancy at the unit level and another is redundancy at the systemic level. Let's take the swarm as a system, in that case if you have multiple backup copies of machines adept at performing particular tasks then you essentially have redundancy provided that the same method is not followed at the unit level. In that case if one failed for a particular set of inputs (hardware or software) then you can "patch" the rest by either avoiding the physical situation or changing the software. You can repeat that with individual units and let them have free reign. If one of them gets destroyed then the other ought to be able to be modified in time. Although it won't guard against stupidity at the unit level, this should to be a much more redundant system than anything we can create within a single one-use device. At the level of the single unit, systematic failures come into play and more copies is actually less over there. Things must be asymmetrically designed to overcome edge cases and systematic failure modes and robust engineering makes sense over there. I think that one of the best ways to implement such a system would be to spend most of the resources in creating and testing a mobile base which is independent of its payload. (attachable modules which perform specific tasks) You should then be able to deploy this machine across missions and learn from all of the real world testing in each mission to create something truly robust and reliable. Once you achieve that you can start offering redundancy at the swarm level through the payloads. For example in that high gain, low gain antenna scenario, wouldn't it be better to have a dozen robots equipped with a variety of antennas dedicated to communication? >>> It's when the designs and tests don't match reality is when the problems happen. <<< Yes, but isn't the entire point of the exercise to fail early and fail often so that you can succeed? If your system is disposable then any systemic failure becomes yet another data point to engineer against and all future systems are better because of it. There is no better laboratory than nature and surely this is a point in favour of it? (unless I'm missing something) Would you like to carry out this convo via email? If so then please feel free to drop me a line at searchingforabsolution [at] hush.com >>> [3] http://www.dau.mil/pubscats/ATL%20Docs/Mar-Apr10/ward_mar-ap.. http://www.dau.mil/pubscats/ATL%20Docs/Mar-Apr10/ward_mar-ap.... <<< Thanks for linking me to this article! It was great.