The Navy wants to teach robots to teach themselves in a learning-by-doing experience

The Navy wants to teach robots to teach themselves in a learning-by-doing experience

Best listening experience on Chrome, Firefox or Safari. Subscribe to Federal Drive’s daily audio interviews at Apple Podcast or One podcast.

Can robots teach themselves new tricks? In theory they could, according to researchers at the US Naval Research Laboratory. In a new white paper, they show how robots, like people, can learn using an educational curriculum and agenda. To see how they would test this theory, the Federal Drive with Tom Temin I spoke with research scientists Laura Hiatt and Mark Roberts of the US Naval Research Laboratory.

Tom Temin: Good. So tell us first of all, before we get into the topic of what you do with robotics, you can make a strong distinction in this document, between curricula and learning agendas. And that’s something that federal humans are currently dealing with with an educational agenda to improve the customer experience and all of this, what’s the learning agenda? How does it differ from the curriculum?

Laura Hyatt: So that’s an important distinction in our methodological work as we define it, it’s similar to what we all encounter when we go to, say, school. Therefore, a teacher, coach, or other professional determines the order in which we learn things. Usually this is from easy to difficult, for example, scaffolding of skills that lead to a gymnast learning a back lip or learning a new sport like racquetball. This order of learning things allows the students to quickly ramp up their skill set and learn what they need to know to achieve a certain level of proficiency such as reaching a certain level of gymnastics, creating an advanced sports team, etc. and a learning agenda, however, when a person decides for himself what he needs to learn and practice to achieve its objectives. So they can do that, for example, based on skills they’re already comfortable with, and then expand from there. And one important implication of having students set their own learning agenda is that they can assess and modify a skill at any time, even in the middle of a practice session, on the skill they are working on next, based on their progress.

Tom Temin: Well, I understand, I’ll say maybe another analogy and you can tell me if that’s accurate. If you’re a pianist, it’s not the syllabus that matters, but how you work on your trills and keep track of your voices and scales. This is your learning agenda.

Laura Hyatt: Yeah, that’s one way to think about it, maybe like a textbook, the right approach can be where you decide the order in which you practice songs and what skills you repeat over and over.

Tom Temin: surely. Well, and even Mac bots. And first of all, what kind of bot are you thinking of to prove it, because there are bots, bots, RPA technology. And all of this, so there are bots in marine contexts walking around with fire extinguishers.

Mark Roberts: These platforms will be straightforward. We’ll start with some kind of research platform as a starting point. The research will focus on a rolling robot first, then one called an extension, and then eventually, on more complex robots.

Tom Temin: These are the kinds of humanoid robots that you see, I mean, robots, sometimes it’s just an articulated arm. And sometimes she walks on her feet?

Mark Roberts: Maybe eventually, in the project, we’ll hopefully get to that point, in the beginning, we’ll use the base type with an articulating arm like you’re talking about.

Tom Temin: Well, then tell us how the learning agenda might work with a robot that has a brain in the sense of a bunch of microprocessors, but isn’t really a thinking machine.

Laura Hyatt: That’s kind of the gist of the project. So one of the ongoing challenges in robotics, sort of, as you alluded to, is the high cost of programming robots to generally be capable. And this project is going to help overcome that by letting the robots, as we said, perform a kind of self-learning to help themselves become more capable. And so the first part is to let the bots set their own learning agenda, like what we said above, the bot can start with what it knows how to do, however, that’s represented in its computer, and kind of work in a way that’s harder and harder the tasks adapt as they are to achieve maximum learning. This includes the robot’s ability to identify tasks that are more difficult than one another. But one of the new and exciting parts of this project is that we also use objectives as part of this learning structure, as well as going from kind of easy tasks to hard ones. For example, a robot performing a task on its own, such as picking up a cup of coffee, might look different than a robot performing a task followed by another action in serving another person. So consider picking up a cup of coffee and then handing it to someone else. So technically you’re doing the same task, you’re just picking up your cup of coffee, but the goal is different. This makes it a little more difficult. So one of the key things here is that we use goals to structure learning by recognizing that goals can and should influence the learning agenda as well.

Tom Temin: We speak with Laura Hiatt and Mac Roberts, two scientists at the US Naval Research Laboratory. What I’m hearing is that there is an implication that there is some artificial intelligence, or even robotic process automation within a bot. And is this something new in robotics that it can teach itself and change what it does over time?

Laura Hyatt: Yes, there is a great intersection between bots and artificial intelligence when we talk about some sort of metaphorical brain of a robot. So definitely, the techniques that we’re going to use here are machine learning, goal reasoning, which is kind of automated planning. And so these are the known AI technologies that we’re going to leverage to use in our bot.

Tom Temin: And tell us about the experience, the specific tasks that you plan for the robots.

Laura Hyatt: So we’ll focus on starting with opening the door. So this is a seemingly simple task, especially with people because people are so adept at navigating around the world. But if you think about it, the doors have a lot of different handles. Some doors don’t have any handles, some doors, some doors push, pull, there’s a lot of variety out there. Suppose for example, we have a robot that knows how to open a swing door. So maybe then, he might learn to open a door where you have to push a bar to get through, that would be kind of the next level of difficulty. Or maybe you’ll try one next with a handle where you have to use your thumb or some other digit of the robot to push the top down to open it. This is a kind of moving through doors of different levels of difficulty. And then that can include opening the door for different goals. So open the door, just check if it’s locked maybe it’s a simple target for that. Perhaps even more difficult is opening up for someone else to walk through. Then, it can be more difficult to keep the door open while holding something fragile or spillable as the robot passes by. So builder it learns. So she frequently makes more complex doors, and for more complex goals until she generally masters the skill of opening doors.

Tom Temin: How about the idea of ​​grabbing the CAC card to open the door and then knowing it’s now unlocked, now you can push the crash bar?

Laura Hyatt: Yes, that might be the gold standard out there. What we should see so far The robots do not have CACs, however.

Tom Temin: Well, this is something that you’ve developed in collaboration with the Naval Research Laboratory, NRL, do you have funding for this and where does that come from?

Mark Roberts: So this is a three year project. It is funded by the Office of Basic Research as part of the LUCI program. This is LUCI stands for University Laboratory Collaborative Initiative. The purpose of this program is to bring together national laboratories and researchers at the best universities. So for this project, we’ll be collaborating with two leading professors of AI research. Professor Dana Nau at the University of Maryland is an expert in automated planning, she has co-written two textbooks on the subject and has led efforts to develop the kind of planning technology that Laura talked about a moment ago, that we will be using in this project. And we’ll be working with Professor George Conan Doris of Brown University, who is an expert in reinforcement learning and has been working on developing techniques for learning abstractions that facilitate rapid learning of actions and cognition.

Tom Temin: And did you build some kind of lab with lots of doors that the robot can walk through using the camera?

Mark Roberts: Not yet, we’ll do that as part of the project. Should start, hopefully in December. But yeah, that will be part of the project. There are some physical places in the NRL where we already have some researchers doing similar pieces to those with other types of doors or gates, so hopefully we’ll be able to build on the work they’ve done.

Laura Hyatt: yes. And it turns out that even in our offices, there are a lot of types of doors to start with.

Tom Temin: Yes that’s right. And maybe you can learn to jump through hoops after that, you know, and then cut the red tape, and you’ll have a really good government bot. But let me put this to you. Are there ramifications for future applications in opening the door? Or is opening the door simply the kind of problem that can really test self-learning theory? And then you can apply it to other activities? Or maybe a little of both?

Laura Hyatt: Yes, it’s the second. So now we’re just interested in the general ability to use the learning agenda and opening the door is what we’re focused on to really kind of figure out what that looks like and what it ends up with. Of course, having a bot that can learn to open all kinds of doors is also an ability that many bots can benefit from as well.

Tom Temin: Well, this research really opens doors on so many levels.

Laura Hyatt: Yes, you can say that.

Tom Temin: Good. Well, I’d like to come see it in action one of these days. Laura Hiatt and MacRoberts are scientists at the US Naval Research Laboratory. Thank you very much for joining me and good luck.

Laura Hyatt: Thank you. We are glad to be here.

Mark Roberts: And thanks for having us here. We appreciate your interest in the project. And we just want to thank our sponsors the Office of Basic Research who is funding this as well as the other sponsors who led the research that led to this project, namely the Air Force Office of Scientific Research and the Office of Naval Research.


#Navy #teach #robots #teach #learningbydoing #experience

Leave a Comment

Your email address will not be published. Required fields are marked *