The thing is, iterators are presented as the preferred alternative to container indices or pointer incrementation over a container. Both of which are modification safe. So you either need to have them match the capabilities, or accept that they're not an alternative.
Nothing unusual there, right? std::map iterators are non-invalidating and you're not touching "iter", so it should be (and is) remaining the same. But what if we use reverse iterators (and correspondingly switch the side of the iterator we do the insert on)?
Despite sounding like the same sort of thing ("iterators"), forward and reverse operators have some very different properties in a key area. A non-invalidating forward iterator will always remain pointing to the exact same element regardless of what happens with other parts of the container. This does not apply to reverse iterators, as they are implemented in a rather unintuitive way - they actually point to the element after the element that they pretend to point to, and so changes to other elements can change what they appear to point to.
There's a lot of things like this in C++ that can slip past a person for years before it actually bites them. Don't get me wrong, I love C++ and think C is a rather dangerous language (from a memory safety standpoint) that requires that its authors reinvent the wheel over and over again. But C++ does have some weirdness in places that can pose hazards. For example, from a more beginner-perspective, what percentage of users have at one point been frustrated by trying to understand why a pointer in one of their classes is getting freed unexpectedly, due to not realizing the dangers of the implicit copy constructor/assignment operator when it comes to pointers? I bet that's bit almost everyone at some point in their career. Sure, you can "reason out" that that would happen, but most people learn it by being bitten once or twice.
Scramjets are rather difficult to orbit. But they are rather interesting for high speed suborbital flight. They let you go through some very thin, high air, very fast. It's also possible to use them for launching into even higher parabolas where they can't fly steady-state, but just drift with nearly no resistance on their way to the destination.
Unless people start making ballistic missiles that look like airplanes, they're going to have very different radar signatures.
If a country doesn't trust another to not hide their missiles as aircraft, then they won't allow ballistic aircraft transit with that country. The US could still allow ballistic aircraft with, say, Britain while not allowing it with, say, Russia. To get permission to launch, every flyover state would have to give permission - which they wouldn't give if they had fears of a concealed nuclear attack.
The reason that a LEO satellite goes 7,8 kilometers per second at no incremental energy cost is because it's in really, really, really thin air. Which is the whole point I've been making in this entire thread: the higher you go, and the thinner the air, the less energy it takes to move through it. You don't have to go all the way to LEO, but every kilometer you increase in altitude cuts your energy consumption significantly.
And really, " Otherwise you have no reason to worry about ceramics, and can work with tin foil." - you've never heard of reentry? One of the biggest heat-challenges confronting modern engineering? Objects moving very high and fast, even if they experience little heating in-transit, experience much on return.
In the atmosphere, you can travel higher, where there is a thinner atmosphere, and you have less drag, but also less dynamic lift.
Which is why for an aircraft wanting to go faster you do one of two things, depending on how much faster we're talking about and the propulsion system:
1) Cruise at a steady higher altitude at the higher speed, such that drag and lift force remains the same - but since you're going faster , you're spending energy to overcome the drag force for less time.
2) Use your in-atmospheric propulsion and lift during ascent to arc you into a parabola. At the peak of the parabola you may have insufficient thrust or lift to maintain a steady state, but the point is not to maintain a steady state.
#1 is more optimal for the lower end of supersonic flight, #2 toward the higher end.
According to you, the flight takes 5 minutes and needs one drop of kerosene now?
Where did I say "20 orders of magnitude difference"? I just said your claim of "a few percent" is not at all representative.
Let's look up some stats about the original 747-100 and a new aircraft (say, the A380). On all comparisons I'll put the original 747 first and the A380 second
Price (2014 dollars): $168m/$428m
Range (same cruising speed): 9045km/14800km
Passengers (three section seating): 397/555 (with more area per passenger)
Maximum load: 142 tonnes/283 tonnes
Operating costs: Difficult to quantify, but the A380 is said to have "direct operating costs per seat 15-20% less than those for the 747-400", which is itself several generations after the original 747-100, so somewhere in the ballpark of 30-40% better is probably a reasonable guess.
Fuel cost per seat per mile: $0.138 / $0.108
Standard features list: Boeing 747-100: Lavatory, Phone System, Reading Light(s) Airbus A380: Bar, Cabin Lighting, Chairs: Berthing, Chairs: Moveable, Galley, Hot Water, iPod Connection, Microwave, Monitor(s), Separate Air Outlets, Soundproof Cabin, State Room, Stowable Work Table, Lavatory, Phone System, Reading Light(s)
Accident history: 3% of all 1505 747s (whole category, not just 100s) ever built have been lost to crashes. Concerning the 747-100,200,and 300 series specifically, the rate is 1.06 fatal crash per million departures. / Despite being produced for over a decade with 173 in service, there's not been a single hull loss or fatal accident with a A380, only one emergency of significance. If we want to compare to a slightly older aircraft to get more miles under its belt, the 747-400 (1988) had a rate of 0.07.
Further detail on accident histories: while it's hard to do a precise safety comparison between the two due to the different service histories, it's worth noting that the biggest advance in the airline industry in the past four decades has been in safety in general. New regulations (resulting in frequent modifications and retrofits, change in design processes, etc) have cut the rate of fatalities per million departures from around 40 when the 747-100 was launched to around 10 today.
The price of aircraft has gone up - but the TCO has gone well down, particularly because of the 30-40% improvement in fuel economy. This has also translated into much greater range for the same general plane configuration.
Your 20 orders of magnitude nonsense? Of course not. But "a few percent"? It's way better than that.
It should be added that, again, the Concorde (ignoring Russia's ill-fated effort) was the first supersonic passenger jet - a prototype, if you will. The 747-100 certainly was not. If you want to make a more proper comparison, you need to compare an A380 with a de Havilland Comet. The comet, by comparison to a passenger jet 40 years later, was a terrible, expensive, inefficient, unsafe plane. But it broke the ground for others to follow.
Again, not necessarily. Your average LEO satellite for example goes 7,8 kilometers every second at no incremental energy cost. Yes, it cost a lot to get it there, but the amount of energy it consumes per kilometer traveled as a whole is miniscule. The fact that there is almost no resistance there is a really big deal. You cannot simply say "because something is going faster then it's using more energy per kilometer". That's a rule of thumb but certainly not a hard, fast rule. When the forces you're experiencing change as you move into a different envelope, your energy consumption changes. And that change can be greater than the energy you burned to get into the different envelope.
It's not that simple with air transport because there's another factor - altitude. Altitude means thinner air, which means reduced air resistance. It's not like transport on the surface where you're facing wind or water resistance roughly proportional to the square of the velocity, there's more factors involved.
Let's not hate on Concorde for too much - it served for 27 years, which while certainly not record-setting wasn't a bad run. British Airways was said to have turned a profit on their runs. It was doomed by a collection of factors - the only crash of the plane in 2000 (not the design's fault, it hit debris shed from a DC10 that the airport should have cleaned up), the 2001 terrorist attacks, and a general downturn in aviation and reduced profits. Also when the plane was grounded after the 2000 accident it's said that the airlines realized that they made more money in shunting their concorde passengers into first class of their other flights - it's not like they had another supersonic plane they could just switch to.
Supersonic commercial travel will certainly happen again, and the next plane will be improved over Concorde in every regard. It really just needs a sustained upswing in long-distance commercial air travel, particularly the high end of the market. Maybe emerging markets will be the spark that's needed.
Your claim about "ceramics" in general is not true. Ceramics are a broad category with widely varying response to heating. Many are perfectly stable with numerous quick heating/cooling cycles (also note: reentry is not a "quick" process from a materials standpoint).
Ceramic parts are widely used in tasks that deal with extreme temperatures, particularly in oxidizing environments. At some temperatures they're really the only things that can take the heat.
It's not even that simple.... in the exosphere, it's more like individual particle collisions than like dealing with a bulk gas. And then when you get deeper you get into the atmosphere, it still doesn't behave like a normal gas - the dense compression shocks of air that you've built up in front of you that are so hot that you actually lose some of the heating energy to endothermic chemical reactions - there's a different equilibrium there than at lower temperatures. While there's enough time to reach the new equlibrium in the shocks in front of the spacecraft, in the sidestream the gas moves past so fast that it doesn't have time to reach its new equilibrium as it cools ("frozen reactions") - the reactions happen at a point well behind the spacecraft, releasing the energy there. So the spacecraft actually gets away with bypassing part of the energy it's losing to the atmosphere.
On the other side, these frozen reactions have downsides too - it's part of what makes scramjets so difficult (the desired combustion being "frozen" to past the end of the craft due to insufficient reaction time). In fact, if this didn't happen, you could potentially make spacecraft that propel themselves in the outer reaches of the atmoshere/low Earth orbit (anywhere over 100km really) without need for onboard propellant by recombining the free oxygen radicals that dominate there. (technically you still probably could, but it would require a long spacecraft indeed)
The point isn't to stop them from getting hot - the point is to not melt or weaken when they do. Ceramics are the best materials in existence for this. For example, hafnium nitride carbide melts at 4126C. Iron boils at 2826C. And this is more meaningful than it sounds - because the only ways during reentry that one can get rid of heat are storage, ablation, and radiation. Depending on the Cp scaling factor, ablation and storage are proprortional to the temperature to the 1-2 power while radiation is proportional to the fourth power of the temperature. So being able to tolerate a given amount of higher temperatures translates to being able to dissipate far greater amounts of reentry heating.
The fact that their first material was silicon oxynitride I find interesting. I don't know how thick their layers are and whether they're able to get any transparency out of them, but thin films of silicon oxynitride are sometimes used for gradient-indexed optics - by changing the ratio of oxygen and nitrogen you can greatly change the refraction index, and thus make things like perfectly flat, thin transparent objects that function as lenses - like a fresnel lens but without roughness or distortion. And when you dope silicon oxynitride you can make phosphors of various colours. So depending on what blend of powder they lay down with the print head they may be able to use it as a rather nifty optics-printer. And since they're using UV to solidify the substance they're basically doing photolithography, aka they should be able to do very fine details. And it's a dielectric with good thermal conductivity. See where I'm going with this? Literally printing your own displays.
According to TFI, their sensors outperform human touch. Also, as per your link, a cheap uncooled PCB "RGB sensor" is not exactly state of the art in spectral analysis, to put it mildly. In the extreme case we can count individual photons at each of highly specified, narrow frequencies with virtually no error. The eye is quite good, but it's not that good.
The whole point of the article is not that they're changing industry standards. The whole point is that they're making a new standard on properties that either have no standard or poor, little used standards at present.
As it stands, while most industries feel comfortable for example simply specifying a color of a product based on a pantone, they insist on samples or travel to ensure that textures come out right. if you wanted to specify how an object should feel, there was no reliable standard way to do that. You could use descriptive words or analogies, make up your own set of measurement standards that you want it to meet, etc... but there was no easy standard way to describe in absolute, measurable terms how things feel. This team's research basically makes a pantone system for texture. The sensors sense each property to higher resolution than a human can, and thus one can be confident that one surface created to specific syntouch specification will feel exactly the same another, one can interpolate properties between different syntouch specifications ("I want a sheet with a texture halfway between these two"), and so forth, just like one can with pantones.
It really depends on the "small object", my experience has been in general less than stellar - ~60 perfectly positioned all-angles images of objects in ideal lighting with the object and nothing in the area (except for me and the camera) moving to get a model with holes in it, and only that if the scene is ideal - nothing too shiny, nothing clear not too complicated, the camera shadow not messing anything up, no inadvertent breezes, no objects that for some reason or another don't happen to meet the software's expectations, no taking odd numbers of photos on the first Tuesday in December, or whatever it is that happens to just ruin them out of the blue.
Yes, my experiences with a variety of them have been less than stellar... and I'm not even talking about the buggy interfaces (don't even get me started on, say, 123D catch... or the numerous ones that don't work at all)
As for "the human brain", I think you're overthinking things. I'm not talking about anything that involves "knowledge", like "I can see one half of a person's head, I'm going to assume symmetry and fill in the other side" or "I can see what looks like a tree here, I imagine there's going to be some branches, leaves etc on the other side where I can't see them". I'm talking about simply raytracing - the ability to go beyond some point matching, to add in how reflection and transmission effects the scene. First low resolution, then at higher resolutions as everything coalesces. The basic concept is not rocket science, and it's certainly not AI, anything related to what the "human brain" has to do - just try a change (based on what worked best last time), render, compare, and see if it made the image better. However, it is a lot of work - both in terms of writing an initial implementation, and then optimizing it so that it works right with widely varied datasources, making it fast and reliable and not prone to converging to suboptimal local minima, etc.
Getting that base point is potentially a 2026 challenge. After then there's no shortage of "human brain" style improvements one could add to it, of course. "Hey, this object that you saw moving from Point A to point B between times T1 and T2... I recognize that as a human, and thus I'm going to fill in all sorts of details that you weren't able to pick up, and furthermore rather than just sliding them I'm going to make their legs move and articulate their whole body in a realistic walking motion, unless they were going over X speed, wherein I'll have them run, and hey, here at T3 they're oriented toward other objects that I identify as humans, and I identify signs of amicable conversation, so I'll have them gesticulate in a realistic manner..." That sort of thing is a whole different ballgame.
There are some "virtual cost" proposals out there, along the lines of proof-of-work problems for delivery of messages from people not a whitelist. In such a case I hope that the standard would recommend having a "send and whitelist" button right next to the send button in email clients to encourage people to whitelist people that they plan to exhange emails with, and/or autowhitelisting after a given number of exchanged messages. Whether a client has whitelisted someone should be transmitted in the header data, so that legitimate mass-mailers (mailing lists, website notifications, etc) know immediately during the subscription process that the user whitelisted them when they sent that "subscribe" email. Another useful item would be a special link format that mailing lists can use in the body such that when the user clicks on it it prompts asking whether or not to whitelist the source. That is to say, the current "Click on this link to complete your subscription" that takes you their website will first involve your mail client prompting you about whitelisting, then will actually follow the link.
The choice of proof-of-work algorithm to limit the rate that non-whitelisted emails can be sent out can be compute problems (ala hash cash), though ideally more memory-lookup time than compute, since it improves slower and isn't as easy to "supercompute". Another possibility is having third-party servers required for providing part of the data on challenge-response problems, causing network or imposed latency onto the process - although somebody would have to pay for said servers. A third possibility is a turing test-based currency to pay for emails (ideally well less than one test per non-whitelisted email, but nonetheless sufficient to interfere with bulk senders). Spammers can use botnets against compute challenges and "mechanical turk" type services against turing tests, but neither of these things come for free. And the whole point is to cut into the margins.
As a fourth option, one could also simply allow anyone who doesn't want to pay compute/turing test costs to simply buy non-whitelisted mail sending rights with a small amount of actual money... it wouldn't require much, as the point is only to hit the margins of those sending millions of non-whitelisted mails per day. Other things that could be considered as worthy of something to "pay" for sending emails would be limited-supply elements that - even where they're traded illegally - would still impose costs to spammers. The above "it only costs money if flagged as spam" idea isn't bad in this regard
I kind of like the idea of implementing all four simultaneously and letting senders choose - the "tickets" needed for sending non-whitelisted emails being gotten for completing memory-intensive compute problems (ideally ones that work on solving real-world problems and doing some good) or turing problems (re-captcha style, so you're doing some good), plus there would be imposed rate-limits and network delays on serving the turing problems and data for the compute problems (with the server costs being borne by those who want peoples' computers solving their problems) and the option to (cheaply) buy tickets with money. And then, as the GP recommended, tickets could ideally be set up to be automatically refunded by the user not marking a message as spam within a certain time period after reading it. Altogether you'd be making life miserable for spammers, doing good, and minimizing any inconvenience to users. The same ticket system could be used by any other system too, not just email - people, say, registering new social media or gaming accounts, sending facebook messages or friend requests to people they're not already friends with, etc.
... photogrammetry (aka conversion of 2d images to 3d data). That would allow any recorded imagery (photos, film, etc) to be experienced in VR from any angle as well as the easy digitization of any object. There exists photogrammetry software today, of course, but it's weak sauce - some variants fundamentally require knowledge of positioning and/or orientation, all have trouble with reflection (including specular reflection), translucency/transparency, shadows, any form of movement, etc, and even in perfect circumstances often do a poor job.
That's not to say that point matching, as is used today, is useless; it's good to help establish initial conditions, particularly if the source data is missing any sort of GPS/dead reckoning positioning/orientation data, and to get an initial rough layout of the geometry and texturing. But it's just not enough on its own - it's only a first step. There's far more to how things look in 3d than just positioning flat static matte opaque surfaces in flat lighting. You have to next proceed to actual renderings and optimize these other properties from there.
What we really need is subsequent steps built around a fast, physically-based raytracer, with each image being treated as an object with optimizeable properties (geometry, maps for color, roughness, reflectiveness, transparency, etc). Discontinuities (say, where a foreground object ends and a background object begins) turn out to not be an actual problem because such an optimization problem would turn the discontinuities transparent (so that they don't interfere with the scene as viewed from other angles) and they'd effectively go away. Similar areas (geometry, location, surface properties) get automatically grouped into new objects. Everything in the scene is a parameter that can be tweaked, from camera positions/imager properties all the way to the light pattern from the (unseen) sky sphere - and the effects of the tweaks evaluated by quickly re-rendering the scene from the point of view of a number of nearby (temporally and geographically) camera positions. Areas where tweaks provide significant benefits are the focus of further tweaks and tesselation (both in terms of geometry and surface properties). Areas providing little benefit become deemphasized and potentially collapsed.
Optimization problems require error minimization. The error is minimized where every part of every image maps has as much mutual mapping as possible with other images, with as little error in the resultant images rendered from each viewpoint, with as few objects as possible and as simple of geometry/surface maps as possible.
There should in theory be two sources of error remaining between the renderings and the real-world images: unseen objects that have an influence on the world (shadow-casters, lights, reflections, etc), and objects whose parameters shift with time. So all error remaining after the previous details are optimizing can be attributed to these things. The possibility of unseen objects can be trialed by treating patches of surfaces with high error as new camera positions and creating the error as a new heightmap-image object to be optimized, just like any other picture. It can either be in front of it (reflection) or behind it (transparency), the latter which also gives the potential need for optimizing refraction parameters (the default scenario should assume no refraction).
Time-related errors can be handled by experimentally tweaking any object's parameters over short periods of time and seeing if the overall errors converge or diverge. The more that transforming an object tends to minimize error, the more likely that further optimization attempts should try transforming it again in different timeperiods. Transformations at a minimum can include translation, rotation, creation, deletion, and surface property changes, but depending on the object model may include things as complex as armature generation/rotation and deformation.
It's a huge task. But I think some day we'll get there. And then a
I suck at welding too, although I think my 1960s MIG welder is partly to blame. Or at least that's my excuse;) And I have to admit that I too have some amateur rocket concepts myself that I'm really itching to try, though I need to retire a few existing projects first. I've got one rocket concept that I'm working on simulating in OpenFOAM involving a caseless LOX/aluminum/paraffin/polyurethane rocket that burns itself up in its entirity. First the channels of open-cell polyurethane foam saturated with LOX burn, then paraffin surrounding them (while their honeycomb aluminum substrate provides compression), then melting and atomizing the aluminum into the exhaust stream (where it provides extra heat to the exhaust), with the exhaust expanding via a virtual expansion nozzle (the point of the cfd simulations is to figure out what geometry yields optimal expansion while not complicating manufacture). It'd be really neat to know if it works out in the real world. But I need to finish the artsy gate for my land that I've been working on first at the very least, and probably a few other projects too, I can't keep accumulating them whenever something else jumps into my head;)
Don't get me wrong, I love coding. But you're absolutely right that there's something particularly special about making something that you can touch and interact with:) And (with the exception of welding well) I usually find that things are easier than they sounded at the outset. For example the other day I was restoring a late 1800s pressure gauge and almost didn't even try to get the dents out of a brass ring because I "don't know how to do bodywork". Turns out that it's no harder than just finding any flat-tipped metal tool, a hard surface, and making a bunch of little taps on the tool with a hammer as you move the tip around - after polish you'd never know it had once been banged up.
Note that that's "most" but not "all". There are of course laser sintering systems, but prints from them are usually very expensive because they're very expensive. Also laser spraying / thermal spraying systems, but they're new and rare at this point. Another technique sometimes used is printing out a model comprised of metal grains inside a plastic matrix, heating up to sintering temperatures leaving a porous version of the object made of the desired metal, and filling in the pores with a lower melting point metal. Quality usually isn't as good with this technique, but it's cheap. That said, lost wax casting does appear to be the most common that they use. And the results are basically just like any other cast object.
Most professional 3d printed metal services actually 3d print molds and then do lost wax casting. They also usually do finishing work too (metal or not), such as sandblasting and the like. The resulting quality is generally superb.
There's a big difference betseen $20-30k (which I don't have) and a few hundred dollars + a pickup (which I do have).;) I don't have personal experience with CNC mills but I am the sort of person who regularly takes up projects building / modifying things (often metal) by hand, and I'm a programmer, so given the two I doubt I'd have trouble learning.
I'd really like to buy an Epilog laser but the ones that I'd want are about $26k, which is pretty expensive for a machine that isn't going to be used to make money
Then make money with it. I recommend 500 yen pieces - they're only 7 grams each (a bit more than a quarter) and worth nearly $5. The right mix of melted down pennies, nickels and copper scrap should give you the right alloy. In Japan you can buy almost anything in vending machines, so just make sure your coins past the electrical conductivity test and you're golden.
Slashdot Mirror
The thing is, iterators are presented as the preferred alternative to container indices or pointer incrementation over a container. Both of which are modification safe. So you either need to have them match the capabilities, or accept that they're not an alternative.
Careful, C++ has its own share of unintuitive weirdness. Take the following example:
This prints out:
Nothing unusual there, right? std::map iterators are non-invalidating and you're not touching "iter", so it should be (and is) remaining the same. But what if we use reverse iterators (and correspondingly switch the side of the iterator we do the insert on)?
Despite sounding like the same sort of thing ("iterators"), forward and reverse operators have some very different properties in a key area. A non-invalidating forward iterator will always remain pointing to the exact same element regardless of what happens with other parts of the container. This does not apply to reverse iterators, as they are implemented in a rather unintuitive way - they actually point to the element after the element that they pretend to point to, and so changes to other elements can change what they appear to point to.
There's a lot of things like this in C++ that can slip past a person for years before it actually bites them. Don't get me wrong, I love C++ and think C is a rather dangerous language (from a memory safety standpoint) that requires that its authors reinvent the wheel over and over again. But C++ does have some weirdness in places that can pose hazards. For example, from a more beginner-perspective, what percentage of users have at one point been frustrated by trying to understand why a pointer in one of their classes is getting freed unexpectedly, due to not realizing the dangers of the implicit copy constructor/assignment operator when it comes to pointers? I bet that's bit almost everyone at some point in their career. Sure, you can "reason out" that that would happen, but most people learn it by being bitten once or twice.
Scramjets are rather difficult to orbit. But they are rather interesting for high speed suborbital flight. They let you go through some very thin, high air, very fast. It's also possible to use them for launching into even higher parabolas where they can't fly steady-state, but just drift with nearly no resistance on their way to the destination.
Unless people start making ballistic missiles that look like airplanes, they're going to have very different radar signatures.
If a country doesn't trust another to not hide their missiles as aircraft, then they won't allow ballistic aircraft transit with that country. The US could still allow ballistic aircraft with, say, Britain while not allowing it with, say, Russia. To get permission to launch, every flyover state would have to give permission - which they wouldn't give if they had fears of a concealed nuclear attack.
The reason that a LEO satellite goes 7,8 kilometers per second at no incremental energy cost is because it's in really, really, really thin air. Which is the whole point I've been making in this entire thread: the higher you go, and the thinner the air, the less energy it takes to move through it. You don't have to go all the way to LEO, but every kilometer you increase in altitude cuts your energy consumption significantly.
And really, " Otherwise you have no reason to worry about ceramics, and can work with tin foil." - you've never heard of reentry? One of the biggest heat-challenges confronting modern engineering? Objects moving very high and fast, even if they experience little heating in-transit, experience much on return.
Which is why for an aircraft wanting to go faster you do one of two things, depending on how much faster we're talking about and the propulsion system:
1) Cruise at a steady higher altitude at the higher speed, such that drag and lift force remains the same - but since you're going faster , you're spending energy to overcome the drag force for less time.
2) Use your in-atmospheric propulsion and lift during ascent to arc you into a parabola. At the peak of the parabola you may have insufficient thrust or lift to maintain a steady state, but the point is not to maintain a steady state.
#1 is more optimal for the lower end of supersonic flight, #2 toward the higher end.
Where did I say "20 orders of magnitude difference"? I just said your claim of "a few percent" is not at all representative.
Let's look up some stats about the original 747-100 and a new aircraft (say, the A380). On all comparisons I'll put the original 747 first and the A380 second
Price (2014 dollars): $168m/$428m
Range (same cruising speed): 9045km/14800km
Passengers (three section seating): 397/555 (with more area per passenger)
Maximum load: 142 tonnes/283 tonnes
Operating costs: Difficult to quantify, but the A380 is said to have "direct operating costs per seat 15-20% less than those for the 747-400", which is itself several generations after the original 747-100, so somewhere in the ballpark of 30-40% better is probably a reasonable guess.
Fuel cost per seat per mile: $0.138 / $0.108
Standard features list:
Boeing 747-100: Lavatory, Phone System, Reading Light(s)
Airbus A380: Bar, Cabin Lighting, Chairs: Berthing, Chairs: Moveable, Galley, Hot Water, iPod Connection, Microwave, Monitor(s), Separate Air Outlets, Soundproof Cabin, State Room, Stowable Work Table, Lavatory, Phone System, Reading Light(s)
Accident history: 3% of all 1505 747s (whole category, not just 100s) ever built have been lost to crashes. Concerning the 747-100,200,and 300 series specifically, the rate is 1.06 fatal crash per million departures. / Despite being produced for over a decade with 173 in service, there's not been a single hull loss or fatal accident with a A380, only one emergency of significance. If we want to compare to a slightly older aircraft to get more miles under its belt, the 747-400 (1988) had a rate of 0.07.
Further detail on accident histories: while it's hard to do a precise safety comparison between the two due to the different service histories, it's worth noting that the biggest advance in the airline industry in the past four decades has been in safety in general. New regulations (resulting in frequent modifications and retrofits, change in design processes, etc) have cut the rate of fatalities per million departures from around 40 when the 747-100 was launched to around 10 today.
The price of aircraft has gone up - but the TCO has gone well down, particularly because of the 30-40% improvement in fuel economy. This has also translated into much greater range for the same general plane configuration.
Your 20 orders of magnitude nonsense? Of course not. But "a few percent"? It's way better than that.
It should be added that, again, the Concorde (ignoring Russia's ill-fated effort) was the first supersonic passenger jet - a prototype, if you will. The 747-100 certainly was not. If you want to make a more proper comparison, you need to compare an A380 with a de Havilland Comet. The comet, by comparison to a passenger jet 40 years later, was a terrible, expensive, inefficient, unsafe plane. But it broke the ground for others to follow.
The same holds for the Concorde
Again, not necessarily. Your average LEO satellite for example goes 7,8 kilometers every second at no incremental energy cost. Yes, it cost a lot to get it there, but the amount of energy it consumes per kilometer traveled as a whole is miniscule. The fact that there is almost no resistance there is a really big deal. You cannot simply say "because something is going faster then it's using more energy per kilometer". That's a rule of thumb but certainly not a hard, fast rule. When the forces you're experiencing change as you move into a different envelope, your energy consumption changes. And that change can be greater than the energy you burned to get into the different envelope.
Serenity's got more than a few ceramic parts in her. ;)
It's not that simple with air transport because there's another factor - altitude. Altitude means thinner air, which means reduced air resistance. It's not like transport on the surface where you're facing wind or water resistance roughly proportional to the square of the velocity, there's more factors involved.
Let's not hate on Concorde for too much - it served for 27 years, which while certainly not record-setting wasn't a bad run. British Airways was said to have turned a profit on their runs. It was doomed by a collection of factors - the only crash of the plane in 2000 (not the design's fault, it hit debris shed from a DC10 that the airport should have cleaned up), the 2001 terrorist attacks, and a general downturn in aviation and reduced profits. Also when the plane was grounded after the 2000 accident it's said that the airlines realized that they made more money in shunting their concorde passengers into first class of their other flights - it's not like they had another supersonic plane they could just switch to.
Supersonic commercial travel will certainly happen again, and the next plane will be improved over Concorde in every regard. It really just needs a sustained upswing in long-distance commercial air travel, particularly the high end of the market. Maybe emerging markets will be the spark that's needed.
Your claim about "ceramics" in general is not true. Ceramics are a broad category with widely varying response to heating. Many are perfectly stable with numerous quick heating/cooling cycles (also note: reentry is not a "quick" process from a materials standpoint).
Ceramic parts are widely used in tasks that deal with extreme temperatures, particularly in oxidizing environments. At some temperatures they're really the only things that can take the heat.
It's not even that simple.... in the exosphere, it's more like individual particle collisions than like dealing with a bulk gas. And then when you get deeper you get into the atmosphere, it still doesn't behave like a normal gas - the dense compression shocks of air that you've built up in front of you that are so hot that you actually lose some of the heating energy to endothermic chemical reactions - there's a different equilibrium there than at lower temperatures. While there's enough time to reach the new equlibrium in the shocks in front of the spacecraft, in the sidestream the gas moves past so fast that it doesn't have time to reach its new equilibrium as it cools ("frozen reactions") - the reactions happen at a point well behind the spacecraft, releasing the energy there. So the spacecraft actually gets away with bypassing part of the energy it's losing to the atmosphere.
On the other side, these frozen reactions have downsides too - it's part of what makes scramjets so difficult (the desired combustion being "frozen" to past the end of the craft due to insufficient reaction time). In fact, if this didn't happen, you could potentially make spacecraft that propel themselves in the outer reaches of the atmoshere/low Earth orbit (anywhere over 100km really) without need for onboard propellant by recombining the free oxygen radicals that dominate there. (technically you still probably could, but it would require a long spacecraft indeed)
When was the last time your coffee mug melted?
The point isn't to stop them from getting hot - the point is to not melt or weaken when they do. Ceramics are the best materials in existence for this. For example, hafnium nitride carbide melts at 4126C. Iron boils at 2826C. And this is more meaningful than it sounds - because the only ways during reentry that one can get rid of heat are storage, ablation, and radiation. Depending on the Cp scaling factor, ablation and storage are proprortional to the temperature to the 1-2 power while radiation is proportional to the fourth power of the temperature. So being able to tolerate a given amount of higher temperatures translates to being able to dissipate far greater amounts of reentry heating.
The fact that their first material was silicon oxynitride I find interesting. I don't know how thick their layers are and whether they're able to get any transparency out of them, but thin films of silicon oxynitride are sometimes used for gradient-indexed optics - by changing the ratio of oxygen and nitrogen you can greatly change the refraction index, and thus make things like perfectly flat, thin transparent objects that function as lenses - like a fresnel lens but without roughness or distortion. And when you dope silicon oxynitride you can make phosphors of various colours. So depending on what blend of powder they lay down with the print head they may be able to use it as a rather nifty optics-printer. And since they're using UV to solidify the substance they're basically doing photolithography, aka they should be able to do very fine details. And it's a dielectric with good thermal conductivity. See where I'm going with this? Literally printing your own displays.
How is it that there hasn't yet been a single "I for one welcome our new insect overlords" post? Is this not Slashdot?
According to TFI, their sensors outperform human touch. Also, as per your link, a cheap uncooled PCB "RGB sensor" is not exactly state of the art in spectral analysis, to put it mildly. In the extreme case we can count individual photons at each of highly specified, narrow frequencies with virtually no error. The eye is quite good, but it's not that good.
The whole point of the article is not that they're changing industry standards. The whole point is that they're making a new standard on properties that either have no standard or poor, little used standards at present.
As it stands, while most industries feel comfortable for example simply specifying a color of a product based on a pantone, they insist on samples or travel to ensure that textures come out right. if you wanted to specify how an object should feel, there was no reliable standard way to do that. You could use descriptive words or analogies, make up your own set of measurement standards that you want it to meet, etc... but there was no easy standard way to describe in absolute, measurable terms how things feel. This team's research basically makes a pantone system for texture. The sensors sense each property to higher resolution than a human can, and thus one can be confident that one surface created to specific syntouch specification will feel exactly the same another, one can interpolate properties between different syntouch specifications ("I want a sheet with a texture halfway between these two"), and so forth, just like one can with pantones.
It really depends on the "small object", my experience has been in general less than stellar - ~60 perfectly positioned all-angles images of objects in ideal lighting with the object and nothing in the area (except for me and the camera) moving to get a model with holes in it, and only that if the scene is ideal - nothing too shiny, nothing clear not too complicated, the camera shadow not messing anything up, no inadvertent breezes, no objects that for some reason or another don't happen to meet the software's expectations, no taking odd numbers of photos on the first Tuesday in December, or whatever it is that happens to just ruin them out of the blue.
Yes, my experiences with a variety of them have been less than stellar... and I'm not even talking about the buggy interfaces (don't even get me started on, say, 123D catch... or the numerous ones that don't work at all)
As for "the human brain", I think you're overthinking things. I'm not talking about anything that involves "knowledge", like "I can see one half of a person's head, I'm going to assume symmetry and fill in the other side" or "I can see what looks like a tree here, I imagine there's going to be some branches, leaves etc on the other side where I can't see them". I'm talking about simply raytracing - the ability to go beyond some point matching, to add in how reflection and transmission effects the scene. First low resolution, then at higher resolutions as everything coalesces. The basic concept is not rocket science, and it's certainly not AI, anything related to what the "human brain" has to do - just try a change (based on what worked best last time), render, compare, and see if it made the image better. However, it is a lot of work - both in terms of writing an initial implementation, and then optimizing it so that it works right with widely varied datasources, making it fast and reliable and not prone to converging to suboptimal local minima, etc.
Getting that base point is potentially a 2026 challenge. After then there's no shortage of "human brain" style improvements one could add to it, of course. "Hey, this object that you saw moving from Point A to point B between times T1 and T2... I recognize that as a human, and thus I'm going to fill in all sorts of details that you weren't able to pick up, and furthermore rather than just sliding them I'm going to make their legs move and articulate their whole body in a realistic walking motion, unless they were going over X speed, wherein I'll have them run, and hey, here at T3 they're oriented toward other objects that I identify as humans, and I identify signs of amicable conversation, so I'll have them gesticulate in a realistic manner..." That sort of thing is a whole different ballgame.
There are some "virtual cost" proposals out there, along the lines of proof-of-work problems for delivery of messages from people not a whitelist. In such a case I hope that the standard would recommend having a "send and whitelist" button right next to the send button in email clients to encourage people to whitelist people that they plan to exhange emails with, and/or autowhitelisting after a given number of exchanged messages. Whether a client has whitelisted someone should be transmitted in the header data, so that legitimate mass-mailers (mailing lists, website notifications, etc) know immediately during the subscription process that the user whitelisted them when they sent that "subscribe" email. Another useful item would be a special link format that mailing lists can use in the body such that when the user clicks on it it prompts asking whether or not to whitelist the source. That is to say, the current "Click on this link to complete your subscription" that takes you their website will first involve your mail client prompting you about whitelisting, then will actually follow the link.
The choice of proof-of-work algorithm to limit the rate that non-whitelisted emails can be sent out can be compute problems (ala hash cash), though ideally more memory-lookup time than compute, since it improves slower and isn't as easy to "supercompute". Another possibility is having third-party servers required for providing part of the data on challenge-response problems, causing network or imposed latency onto the process - although somebody would have to pay for said servers. A third possibility is a turing test-based currency to pay for emails (ideally well less than one test per non-whitelisted email, but nonetheless sufficient to interfere with bulk senders). Spammers can use botnets against compute challenges and "mechanical turk" type services against turing tests, but neither of these things come for free. And the whole point is to cut into the margins.
As a fourth option, one could also simply allow anyone who doesn't want to pay compute/turing test costs to simply buy non-whitelisted mail sending rights with a small amount of actual money... it wouldn't require much, as the point is only to hit the margins of those sending millions of non-whitelisted mails per day. Other things that could be considered as worthy of something to "pay" for sending emails would be limited-supply elements that - even where they're traded illegally - would still impose costs to spammers. The above "it only costs money if flagged as spam" idea isn't bad in this regard
I kind of like the idea of implementing all four simultaneously and letting senders choose - the "tickets" needed for sending non-whitelisted emails being gotten for completing memory-intensive compute problems (ideally ones that work on solving real-world problems and doing some good) or turing problems (re-captcha style, so you're doing some good), plus there would be imposed rate-limits and network delays on serving the turing problems and data for the compute problems (with the server costs being borne by those who want peoples' computers solving their problems) and the option to (cheaply) buy tickets with money. And then, as the GP recommended, tickets could ideally be set up to be automatically refunded by the user not marking a message as spam within a certain time period after reading it. Altogether you'd be making life miserable for spammers, doing good, and minimizing any inconvenience to users. The same ticket system could be used by any other system too, not just email - people, say, registering new social media or gaming accounts, sending facebook messages or friend requests to people they're not already friends with, etc.
... photogrammetry (aka conversion of 2d images to 3d data). That would allow any recorded imagery (photos, film, etc) to be experienced in VR from any angle as well as the easy digitization of any object. There exists photogrammetry software today, of course, but it's weak sauce - some variants fundamentally require knowledge of positioning and/or orientation, all have trouble with reflection (including specular reflection), translucency/transparency, shadows, any form of movement, etc, and even in perfect circumstances often do a poor job.
That's not to say that point matching, as is used today, is useless; it's good to help establish initial conditions, particularly if the source data is missing any sort of GPS/dead reckoning positioning/orientation data, and to get an initial rough layout of the geometry and texturing. But it's just not enough on its own - it's only a first step. There's far more to how things look in 3d than just positioning flat static matte opaque surfaces in flat lighting. You have to next proceed to actual renderings and optimize these other properties from there.
What we really need is subsequent steps built around a fast, physically-based raytracer, with each image being treated as an object with optimizeable properties (geometry, maps for color, roughness, reflectiveness, transparency, etc). Discontinuities (say, where a foreground object ends and a background object begins) turn out to not be an actual problem because such an optimization problem would turn the discontinuities transparent (so that they don't interfere with the scene as viewed from other angles) and they'd effectively go away. Similar areas (geometry, location, surface properties) get automatically grouped into new objects. Everything in the scene is a parameter that can be tweaked, from camera positions/imager properties all the way to the light pattern from the (unseen) sky sphere - and the effects of the tweaks evaluated by quickly re-rendering the scene from the point of view of a number of nearby (temporally and geographically) camera positions. Areas where tweaks provide significant benefits are the focus of further tweaks and tesselation (both in terms of geometry and surface properties). Areas providing little benefit become deemphasized and potentially collapsed.
Optimization problems require error minimization. The error is minimized where every part of every image maps has as much mutual mapping as possible with other images, with as little error in the resultant images rendered from each viewpoint, with as few objects as possible and as simple of geometry/surface maps as possible.
There should in theory be two sources of error remaining between the renderings and the real-world images: unseen objects that have an influence on the world (shadow-casters, lights, reflections, etc), and objects whose parameters shift with time. So all error remaining after the previous details are optimizing can be attributed to these things. The possibility of unseen objects can be trialed by treating patches of surfaces with high error as new camera positions and creating the error as a new heightmap-image object to be optimized, just like any other picture. It can either be in front of it (reflection) or behind it (transparency), the latter which also gives the potential need for optimizing refraction parameters (the default scenario should assume no refraction).
Time-related errors can be handled by experimentally tweaking any object's parameters over short periods of time and seeing if the overall errors converge or diverge. The more that transforming an object tends to minimize error, the more likely that further optimization attempts should try transforming it again in different timeperiods. Transformations at a minimum can include translation, rotation, creation, deletion, and surface property changes, but depending on the object model may include things as complex as armature generation/rotation and deformation.
It's a huge task. But I think some day we'll get there. And then a
I suck at welding too, although I think my 1960s MIG welder is partly to blame. Or at least that's my excuse ;) And I have to admit that I too have some amateur rocket concepts myself that I'm really itching to try, though I need to retire a few existing projects first. I've got one rocket concept that I'm working on simulating in OpenFOAM involving a caseless LOX/aluminum/paraffin/polyurethane rocket that burns itself up in its entirity. First the channels of open-cell polyurethane foam saturated with LOX burn, then paraffin surrounding them (while their honeycomb aluminum substrate provides compression), then melting and atomizing the aluminum into the exhaust stream (where it provides extra heat to the exhaust), with the exhaust expanding via a virtual expansion nozzle (the point of the cfd simulations is to figure out what geometry yields optimal expansion while not complicating manufacture). It'd be really neat to know if it works out in the real world. But I need to finish the artsy gate for my land that I've been working on first at the very least, and probably a few other projects too, I can't keep accumulating them whenever something else jumps into my head ;)
Don't get me wrong, I love coding. But you're absolutely right that there's something particularly special about making something that you can touch and interact with :) And (with the exception of welding well) I usually find that things are easier than they sounded at the outset. For example the other day I was restoring a late 1800s pressure gauge and almost didn't even try to get the dents out of a brass ring because I "don't know how to do bodywork". Turns out that it's no harder than just finding any flat-tipped metal tool, a hard surface, and making a bunch of little taps on the tool with a hammer as you move the tip around - after polish you'd never know it had once been banged up.
Note that that's "most" but not "all". There are of course laser sintering systems, but prints from them are usually very expensive because they're very expensive. Also laser spraying / thermal spraying systems, but they're new and rare at this point. Another technique sometimes used is printing out a model comprised of metal grains inside a plastic matrix, heating up to sintering temperatures leaving a porous version of the object made of the desired metal, and filling in the pores with a lower melting point metal. Quality usually isn't as good with this technique, but it's cheap. That said, lost wax casting does appear to be the most common that they use. And the results are basically just like any other cast object.
Most professional 3d printed metal services actually 3d print molds and then do lost wax casting. They also usually do finishing work too (metal or not), such as sandblasting and the like. The resulting quality is generally superb.
Your average home 3d print? Not so much...
There's a big difference betseen $20-30k (which I don't have) and a few hundred dollars + a pickup (which I do have). ;) I don't have personal experience with CNC mills but I am the sort of person who regularly takes up projects building / modifying things (often metal) by hand, and I'm a programmer, so given the two I doubt I'd have trouble learning.
Out of curiosity, does anyone ever DIY aluminum extruders? I'm curious as to how hard that is.
I'd really like to buy an Epilog laser but the ones that I'd want are about $26k, which is pretty expensive for a machine that isn't going to be used to make money
Then make money with it. I recommend 500 yen pieces - they're only 7 grams each (a bit more than a quarter) and worth nearly $5. The right mix of melted down pennies, nickels and copper scrap should give you the right alloy. In Japan you can buy almost anything in vending machines, so just make sure your coins past the electrical conductivity test and you're golden.