All right everyone. I am on the line with Ken. Goldberg can is a professor of engineering at UC Berkeley. Ken Welcome to the TWAT podcast. Thank you pleasure to be here. It is great to finally get you on this show. We've been talking about this for a bit. You know I meant to ask you before we started last time you were. You mentioned you. Were working on a book. Maybe we'll get remembering that right. Well I think I'M I. I've been thinking about that for a while but I'm also thinking about an more right now an article. Okay okay. Well we'll We'll get to the article. I think I I came across you and some of your work in the context of decks net. I saw that at a Siemens Innovation Fair last year. I think we exchanged tweets and stuff like that. But you know I would really love for you to introduce yourself to the audience and share a little bit about your background and how you came into working in robotics and okay great. I well first since you mentioned twitter I should mention my twitter handle which is at Ken. Underscore Goldberg. And I've been trained very well. My daughter to post there at least one today so I've got the actually. I found it very interesting channel so so I am posting technical things as well as updates about things that are that are finding out which is the learning about which I find very useful. So my background is that I was. I went to University of Pennsylvania and then went to Carnegie Mellon for I was at USC for four years and then to Berkeley where I've been for now twenty five years for here. I RUN A lab. The we we call it the auto lab for Automation Science and Engineering and we have approximately thirty students doing research in there. And we're doing work. There's there's there's post graduate students and a good number of undergrads and we're also associated with other labs like the Berkeley Research Lab and the rise lab and citrus and other programs at Berkeley our particular labs interested in in in doing research on on robotics basically on algorithm ick approaches to robotics and specifically in last year's been focusing on learning methods for for imitation learning deep learning and reinforcement learning for control of robots in applications from grasping as you mentioned which is a primary want working on for for thirty five years to surgery surgical assistance Hugh assisting human surgeons for for robotics and home robots to especially for seniors and in who are who are who prefer to live at home and the last year is very new and we can talk about later is is agriculture and we have a new approach to poly culture farming that were exploring using deep learning so one thing that I thought was really interesting in looking at your bio is in spite of the fact that you are a highly accomplished robot assists you start your your body starts with Ken. Goldberg is an artist so art clearly must be very important to. You actually saw some sketches behind. You am curious. I'm curious about Ken as an artist. And you know how if all ties into your work. It's not the usual fare of this podcast but then I saw somewhere else. You are filmmaker as well Is that your art? Tell US okay. Well actually I wanted to be an artist when I was a kid and I I basically my mother said listen. You can be an artist after you become an engineer. So She she. She was very wise and I think it was. It was it was a good choice for me because actually love both art. Something that I take very seriously. I think it's often underrated by many people especially Engineers who think of it as as lightweight. It's actually just opposite trying to produce something that's meaningful in the art. World is extremely difficult and demanding. So I've spent a lot of time studying I have made a series of installations and projects. That almost always involve technology in some way. But they're also commenting on the role of technology in society. So probably best known pieces of project is a project called Tele Garden that my students and I set up in the very early very early years of the Internet. So it's nineteen ninety five that we we connected a industrial robot arm to the web interface at the time which was mosaic Browser and we built an interface. That would allow you from your screen from anywhere from your laptop There were no cellphones at the time. But you could. You could log in this thing I think. Yeah it was very fun project. We thought well. It's kind of curious. who would use it if anyone and we got thousands of people coming in and and moving the robot but the part of what was made. An artwork was the context because it was sitting inside a garden. A real physical gardens. We could plant in water seeds remotely and then we got tens of thousands and we estimate that over the time that product was was that robot was available online which is approximately nine years. It was visited over. Hundred thousand people participated in the in the project. That's awesome that's awesome again kind of the technology and are coming together rate. So that was the thing Sam because one of the ideas were said I. I don't think I would have pursued that if I just stuck with my research plans at the time but because this came out and offered a way to reach a at the time when I saw as potentially very broad audience I started putting effort into this then there was a fantastic team of students who worked on it. And then we are thrilled with the the idea that you could take a robot and you could put it into the hands essentially of potentially millions of people and then there were. There was a proof of concept the interface questions there it turned out that there were lots of interesting theoretical questions that came out of that so after that project we did a series of subsequent projects and then had an NSF grant to develop versions of this. We have a patent related to the south. Yeah it really grew into a whole new direction of research that that really started with our awesome awesome into tell us a little bit about your research interests nowadays more broadly. So we're still doing art and I can come back to that. There's a new contract. But the the the lab right now is been been very very focused on robot learning and especially as as I know your. Your listeners are very aware there's been huge revolution in the past decade. And so we've been. We're interested in this before the the advanced in deep learning started but now it really has become a huge focus for us so in particular. We have this been working robot grasping for many years and then went deep. Learning came out. We saw an opportunity to apply it. I can tell you that story if you if you like how we do it. Maybe start from the perspective of the grounding on the challenges associated with grasping like we see these pictures of whether they're rohbock robot hands or more industrial types of robots or prostheses. And you know a can grasp like we've seen we've all seen pictures of that but maybe it's harder than it looks or you know maybe the opportunities that have not figured out. Oh good okay so I can. I can answer that partly. I've realized only last few years that part of the reason I believe I went into this field was at myself as a kid was was incredibly clumsy. I still I still am. Anyone would throw me a ball I would drop it and so You know the last kid getting picked for any sports Games or anything like that and it was just Ed. I think that may unconsciously made me interested in in trying to figure this thing out like how. How do you grasp things and many years? Later when I was in Undergraduate I joined a laboratory of the University of Pennsylvania and they were studying various aspects of tactile sensing and I built a very simple hand with another student and we started really exploring this question. Of How do you grasp things? And it is fundamentally difficult for robots like to say that robots remain incredibly clumsy today. They're much better than they were but industrial arms. If you give them novel objects there will drop them with a fairly high frequency and this is a problem because we really want is want You WANNA be able to pick up anything that's put in front of you and the application the big application that's growing enormous right now is e commerce so you wanna be able to take objects every orders different so you wanna take things from bins and pack them. Lift them out of the band. Grasp and put them into boxes or bags for shipment and that turns out to be a bottleneck right now for robotics