Open Problems in Robotics • Hacker News Board

Re: Open Problems in Robotics

#191

Post by cpgxiii » Thu, Aug 20, 2020, 2:07 AM UTC

dorgo wrote: post #175

YeGoblynQueenne wrote: post #91

Earlier quoted context omitted.

Convex/ non-convex optimisation refers to the shape of the error function we're trying to optimise. In convex optimisation we can assume it's, well, convex: . . ε \ / \ / \ / \ / '._ _.' ^ Global optimum In non-convex optimisation we can't make any assumption about the shape of the error function: ,--. .--. ε / \ / \ ,--. / \ / \ / \ / .' \ / \ / \ / \ / \ / \ / `--' \ / `--' ^ \ / ^ | `--' | | ^ | | | | | | | `-----…

>Generally we prefer to find a _global_ minimum of the error function because then we can expect the resulting approximator to generalise better to data that was not available during training. Sorry to nitpick, but is this true? We are doing optimization here and a global minimum is just a better solution than a non-global minimum. Is there a connection to generalisation here?

In an ML context, the global optima may often correspond to better performance and (hopefully) a more general solution.

In a motion planning or controls context, a local minima can often mean a configuration or path that is infeasible due to collision or wildly inefficient.

Re: Open Problems in Robotics

#192

Post by deepGem » Thu, Aug 20, 2020, 2:28 AM UTC

317070 wrote: post #170

deepGem wrote: post #77

Earlier quoted context omitted.

"The real problem: solve any of these problems, make very little money" - Just curious why have you come to this conclusion ? Object manipulation has potential products in dishwashing and vegetable chopping - sufficiently large markets, potential billion $ outcomes for a startup which takes the early mover lead. Two robotic hands that can work in co-ordination just as human hands do. Extremely difficult to solve, but…

> Two robotic hands that can work in co-ordination just as human hands do. Extremely difficult to solve, but money is there. You will find the most automated Mc Donalds in the world in Switzerland. Mc Donalds over there has a lot more automation there than it does in the US, both in the cooking and in the order taking. Either this is because the automation is extremely difficult to do, but is simpler in Switzerland.…

Can I surmise that most automation in dishwashing etc is geared for commercial enterprises ? I find the same lacking in a household. Sure you have the traditional dishwashers and vegetable choppers but they are largely clumsy to use and still take quite a lot of effort.

For a consumer, I don't know if the equation 'human labor cost << robots' holds true. It is a hassle to get human labor and there is no scope for time arbitrage. A robot can do the dishwashing job at night for example. Unlike a traditional dishwasher, you don't have to load the utensils. Just leave the utensils in a sink and you are done.

Re: Open Problems in Robotics

#193

Post by deepGem » Thu, Aug 20, 2020, 2:32 AM UTC

Nasrudith wrote: post #189

deepGem wrote: post #77

Earlier quoted context omitted.

"The real problem: solve any of these problems, make very little money" - Just curious why have you come to this conclusion ? Object manipulation has potential products in dishwashing and vegetable chopping - sufficiently large markets, potential billion $ outcomes for a startup which takes the early mover lead. Two robotic hands that can work in co-ordination just as human hands do. Extremely difficult to solve, but…

I would have thought the money wouldn't be there for a different reason - human hands suck compared to tools which is why we created them in the first place. It is a bit of a silly trope to have industrial work done by humanoid robots with hand tools. The ability to manipulate destandardized sizes would be useful but precision manufacturing ate most of the lunch long ago which reduces it to more of a "last mile" task…

Well, the advantage I see with humanoid hands is that the same task that a human does can be done much faster by a humanoid hand. For instance chopping onions or other vegetables. Even dishwashing.

Re: Open Problems in Robotics

#194

Post by smilekzs » Thu, Aug 20, 2020, 4:41 AM UTC

jandrese wrote: post #177

Animats wrote: post #173

Earlier quoted context omitted.

The amount of money that's gone into that area without shipping a product is insane.

The potential upside for the company that gets it right is enormous. Billions of people are tired of wasting their time driving. Entire industries can be built on the technology if it works well. That said it's a problem that steers awfully close to needing a full real AI and that's been a showstopper for loads of potential solutions for decades now.

My observation is that the industry ("we") as a whole is slowly but steadily making progress, without throwing our hands and say we need AGI. A big part of this comes from the upscaling of test fleets, which now generate data with enough quantity and variety that you can leverage the existing data tools to analyze and give developers precise and actionable feedback. While Tesla might seem like a popular punch bag in the ADV world, they are a big practitioner of this paradigm and did get good value out of it. That, and the trend of replacing more and more hand-crafted rules with fuzzy numerical models designed to "blend together" the rules (therefore retaining explainability).

Re: Open Problems in Robotics

#195

Post by nojvek » Thu, Aug 20, 2020, 5:26 AM UTC

What really surprises me is Apple is now 2 trillion dollars with another gajillion sitting in cash. Why are they being an optimization company and not really pushing the edge on robotics and automation? I was excited about Apple car but it seems that project is ded.

Why not invest that money in moonshot ideas ? It seems Elon musk is the crazy one with Tesla, Solar City and SpaceX.

It blows my mind how their massive rockets reach orbit and land back. Why haven’t we been making breakthroughs like this in robotics?

Re: Open Problems in Robotics

#196

Post by christopher22 » Thu, Aug 20, 2020, 5:54 AM UTC

Animats wrote: post #66

- Motion planning: already discussed. - Multiaxis singularities: much less of a problem than it used to be. We don't need closed-form solutions any more; we have enough CPU power at the robot to deal with this. You need some additional constraint, like "minimize jerk" when you have too many degrees of freedom. - Simultaneous Location and Mapping. SLAM for short: Getting much better. Things which explore and return a…

Kinda feels like that old life pro tip. Want to know the answer to something? Don’t ask a question, but make a false statement. Every expert in the field will rush to correct you and give you the most up to date and relevant information.

Re: Open Problems in Robotics

#197

Post by foerbert » Thu, Aug 20, 2020, 5:57 AM UTC

m0zg wrote: post #60

foerbert wrote: post #27

Earlier quoted context omitted.

Humans are capable of generating and understanding creativity and complexity that are simply impossible for non-AGI automation. Even then, we don't just let people figure things out for themselves. We put them through training, and then test them. Even after that, we make them liable for negligence. I don't think it's an obvious conclusion that error bounds aren't important for automation because they aren't calculab…

My point is not that it's not important. If someone comes up with a rigorous way to obtain error bars, that's great! I'll take it! My point is that trust should not and will not depend on it. How do you even quantify something like this to a layperson in order to persuade them? Let's do a thought experiment: let's say, we had a self-driving car that's verifiably 10x better than human on average, yet does not provide…

I think a lot of the nuance has been lost here. Even throwing around phrases like "10x better than a human" is already implying a lot of very vague measurement or knowledge. In what ways, and in what environments, is it better?

I've dipped my toes into robotics a few times, and every time I end up being reminded of just how painful it is. Even when you're working in an idealized simulator, it's extremely easy to find a little edge case that causes completely bizarre behavior. And moving out of the simulator only makes things far, far worse.

It's really easy to forget about those sorts of details and brush it away as just things to be solved while developing the automation. But they don't just go away so easily. Error bounds are there to help manage these issues and ensure we know how to best use the automation.

In many instances I think people would tend to prefer the Uber driver in a moderate reading of your scenario. A human driver is likely to perform somewhat consistently and predictably. If they drift around the corner and leave a long skid mark in front of your house you can make some assumptions about how they are going to drive. If you get in and see them struggling to keep their eyes open, you can again make some assumptions about their performance. A well-rested and safe driver is extremely unlikely to suddenly throw themselves into on-coming traffic with no warning. You can refuse or stop the ride if you judge you are not safe.

Automation is a different beast. It's liable to fail in ways that a human driver would not. It may be performing wonderfully until something a human would not even notice happens, and which point it may indeed throw you into on-going traffic. For example, look at adversarial examples in deep learning. As a passenger in this case you don't have a way to judge your own moment-to-moment safety. Even if it is safer on average, the sheer unpredictability and the resulting stress is likely going to eat significantly into any gains.

Re: Open Problems in Robotics

#198

Post by tomp » Thu, Aug 20, 2020, 6:47 AM UTC

jaaron wrote: post #10

fluffy87 wrote: post #7

Earlier quoted context omitted.

This is the scariest part of using machine learning as an engineer on any practical application as well. Without an error bound, ML can’t be in charge of anything that could put human lives at risk. This is also why I don’t understand all the hype about FSD / L5 autonomous driving. We don’t even know yet if such error bounds even exist, so we don’t even know if machine learning is even the right tool for FSD yet. All…

What do you think the error bounds are for a human? I know it sounds like a flippant question, but for certain applications, if we can get a model that's better than human, then it doesn't need to be perfect. And they way we currently do this in all sorts of ways is to pair a human with a computer so that they each do what they're best at. It doesn't have to be about full automation.

This is another one of that "extreme tail risk" scenarios, like climate change and GMOs, that people have wildly different and contradictory reactions to.

Sure, the "legacy" intelligence / climate / food could also have extreme tail risks, it's just that it's been tested for 100s of millenia... whereas new technology might be better in the average case or even 99th percentile, but the 1% (or 0.0001%) is unknown and potentially much worse.

However, it seems to me that people resolve this more along ideological / political lines than with any kind of rational reasoning.

Re: Open Problems in Robotics

#199

Post by fxtentacle » Thu, Aug 20, 2020, 9:14 AM UTC

cpgxiii wrote: post #187

fxtentacle wrote: post #114

Earlier quoted context omitted.

It "just works" for our use case. We put in 4K @ 60fps and receive 960x540 @ 20fps of stereo correspondence pairs. So every matched pixel is averaged over 3 frames in time and 4 pixels in every direction in space, meaning 9x9 convolution kernels. In other words, we make the video super clean by area sampling in space and time. The specific part about our system that makes it usable for me is that for pixels that cann…

Yes, in general if you can capture multiple views you can make up for many dropouts and artifacts. The more you can move, the more likely that you'll get a complete reconstruction. Single-view artifacts are more of a problem when reaching into confined areas or during visual servoing where you don't necessarily have the freedom to move around to get a better view. The results for the Sintel benchmark do look interest…

One trick that I have seen for confined environments is to rotate the plate with both stereo cameras around its forward axis. That way, you can convert the left-right stereo to up-down stereo and especially with highly reflective stuff, there's a chance that that will be enough of a change to reduce reflections.

Re: Open Problems in Robotics

#200

Post by YeGoblynQueenne » Thu, Aug 20, 2020, 10:41 AM UTC

dorgo wrote: post #175

YeGoblynQueenne wrote: post #91

Earlier quoted context omitted.

Convex/ non-convex optimisation refers to the shape of the error function we're trying to optimise. In convex optimisation we can assume it's, well, convex: . . ε \ / \ / \ / \ / '._ _.' ^ Global optimum In non-convex optimisation we can't make any assumption about the shape of the error function: ,--. .--. ε / \ / \ ,--. / \ / \ / \ / .' \ / \ / \ / \ / \ / \ / `--' \ / `--' ^ \ / ^ | `--' | | ^ | | | | | | | `-----…

>Generally we prefer to find a _global_ minimum of the error function because then we can expect the resulting approximator to generalise better to data that was not available during training. Sorry to nitpick, but is this true? We are doing optimization here and a global minimum is just a better solution than a non-global minimum. Is there a connection to generalisation here?

I'ts like cpgxiii says. You're right to nitpick though, because there are no certainties. We optimise on a set of data sampled from a distribution that is probably not the real distribution, so there's some amount of sampling error. Even if we find the global optimum of our sampled data, there's no reason why it's going to be close to the global optimum of our testing data.

But - there are some guarantees. Under PAC-Learning assumptions we can place an upper bound on the expected error as a function of the number of training examples and the size of the hypothesis space (the set of possible models). The maths is in a paper called Occam's Razor: https://www.sciencedirect.com/science/article/abs/pii/002001...

Unfortunately, PAC-Learning presupposes that the sampling distribution is the same as the real distribution, i.e. what I said above we can't know for sure.

In any case, I think most people would agree that a model that can reach the global minimum of training error on a large dataset has better chance to reach the global minimum of generalisation error (i.e. in the real world) than a model that gets stuck in local minima on the training data. Modulo assumptions.