Ken Goldberg
speaker
221 appearances
3 recordings
3 series
first heard Jan 2025
last heard 7 May
Ken Goldberg’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in May 2026 with 1.
Appearances
So I have no idea. But what can we do? We have to say, go ahead. Otherwise, it feels all mapped out, maybe.
Yeah, right. Like the panties on the ground. Right. Exactly. So he drops his shoe into the bin, and we're all sitting there. And the robot just reached over and picked up the shoe.
Wow. And I remember calling Tiffany, my wife, and I said, this was the best moment of my life. Wow. And she said, what about our wedding?
That's exactly right. That's very important. You can't tell it to select a specific object.
You can't say go through this bin and find me all the pennies.
That's called rummaging. Okay. That's very interesting. We're looking at that now. Much different problem, much harder. That could be incredible for recycling. Yes. And also, you think about it, you do this all the time. If you reach in your pocket and you want to pull out a pen, you can always find the pen or your purse.
Right? People are very good at that. And what's going on is very complicated. Robot cannot do that at all. But pulling one thing out of the bin is really interesting because you have to kind of move things around a little bit, sort of see a little piece of it, then pick it up. And that's actually very important for warehouses and for Amazon to be able to deliver packages.
You have to be able to rummage and find the thing you want. And that's still unsolved.
Yeah, because there's physics. We really don't understand friction. And friction is so important. It's what lets us all sit here and things not slipping around. Friction is so important, but it's a very, very complex process. We can approximate it, and there's this model Coulomb friction, etc. But to really get friction right is actually impossible.
If I want to push something across the table, the way it's going to move and react to my pushing force is going to depend on what's underneath it. So if you have one grain of sand... It's going to change it. Yeah, if it's in the right corner, the whole thing's going to rotate clockwise. Exactly. If it's in the left corner, it's going to rotate the other way. But I can't know that.
The robot can't know it. So right there is like one of the great mysteries of nature, right? You don't have to talk about quantum physics. That is one unknowable thing that's sitting right in front of us. Oh, my God. Wow. And we deal with it all the time. So you might say, what do we do? Well, we kind of compensate. When we reach for a glass, we don't just reach our gripper right up to it.
We scoop it up. We're almost anticipating the many different ways it could go wrong. Exactly. We haven't figured out how to do that for robots yet.
It's both. It's largely software because we don't have the sensors. We don't have the control. We don't understand the models of physics. But I also think we need better grippers, too. But that's a whole other story. But the bottom line is that we're far from anything approximating human level performance. And there's been so much hype. And that's what I worry about.
I really do. I think we're on a collision course with a kind of bubble that's going to burst because people are expecting that we're almost there, especially when they see these videos. OK, great. Tell me about these videos because you watch them and you think. We're there. And this is a big problem.
Right. Okay, the first thing to ask is how many takes were required? Many times they get to work once, and that's the video they show.
It's violent when it gets it wrong. In a research lab, that's what we're dealing with. It's always failing. You'd be lucky if you get it to work once. But so if you put it on YouTube, we'll say the success rate is one out of 200 or something. But nowadays, there's so much hype that is not putting those caveats in there.
Someone might say, I have to do that. That's how I'm going to get the next round of funding. But I really cringe about that because that goes against my instincts of I like to say under promise and over deliver. And I'm going to be really careful, never over promise about what we're doing or a result in a paper. We're always careful. Don't exaggerate the result.
And it really is a problem for robotics where you see the videos. And then the other thing is they can be teleoperated. And there's a human behind the curtain. Okay, so what about this Tesla robot?
Okay, so I know this is going to disappoint people when I tell you this, but it's far from being human-like in its abilities. In dexterity, it's very, very weak. It looks good, and there are beautiful designs. And they actually have made progress in the motors and the hardware so it can move more smoothly. And they're also getting very good at walking.
So there's definitely something positive there. But what can they do? And see, this triggers that old idea that we've had in the back of our mind, which is we want these things. We've been reading about them, watching them on movies and TV. We think they're going to come. And yet there's this huge gap. So if you watch carefully at the demos, they're being somewhat teleoperated.
Showing 141–160 of 221 · page 8 of 12
← Previous
Next →