Prism logo

Polar Radar for Ice Sheet Measurements

curved ice line
jagged ice
Home>Prism Update>Summer Seminar - Dr. Agah

Introduction To Mobile Robots

I was told that I have to speak to a general audience and because of that no equations. I've tried to use equations in the past and, if anything, it sort of makes people not want to do it; and I'm trying to encourage more of you to study robotics.

My name is Arvin Agah and I am a faculty with the Electrical Engineering and Computer Science department here at the University of Kansas. (points to slide) That's my e-mail and web address in case anybody wants to talk to me after the talk.


I will try to stay with this outline. I'll try to introduce to you mobile robotics field. I think history is very important. And one of the nice things about artificial intelligence, or robotics, is that it's a brief history. We haven't been doing this for too long.

If you have ever taken math and they say they are starting with history, you can run away because you know it's going to be years and years, but I can cover history of robotics in a very short time.

I'll talk about current status - where we are.

I'll talk about the applications. I have a lot of cool pictures that illustrate some of the ways mobile robotics can actually benefit us or be used.

Challenges. That's the way to look in the future. What are some of the challenges we are facing? What are some of the problems we need to solve to be able to do a better job?

Then I'll talk about trends. And anybody who tells you they are going to talk about the future trends, well, they are really speculating. You don't really know. So I am saying, "Look, I am going to be making some speculations and no guarantees on that."

Then I'll briefly talk about philosophical issues. A lot of you have watched science fiction movies and you know there is a robot that becomes bad and then what do it usually do? It kills everybody in the city and then you have to go into the basement and fight them or whatever. I'll talk about that.


A lot of times when people hear of the term "robot" they think of an arm moving around. This is the field of robotic manipulation. Robotic arms are not mobile. They're still interesting, but it's not an area of research that I work on.

And according to some industry experts, it's an area that's solved. Every one of you who is driving a car right now is driving a car which was probably welded and painted by a non-mobile robot. So they can do very high precision work. If the work is precise enough to actually manufacture stuff, there is no need to go further.

There are companies like Fanuk and Mitsubishi that are building amazing arms that do very cool stuff. But the fact that they are not mobile and the fact that they are not autonomous, in other words, they don't think on their own, makes them different from what we will be talking about today.

If you do study these things, then you spend a lot of time dealing with issues like kinematics and dynamics.

I tried to use this in my lectures in the past in Robotics classes but there is a tendency not to like it as much. It's much more interesting when you build a dog that moves around than we have an arm that goes back and forth and paints things. But this is how robotics got started.

This is a simpler form of robotics because the base is stable. So it's very easy to deal with safety issues. You put a cage around this thing and you are fine. Nothing ever happens. But things change when the robot starts moving around and that's where we want to be.


And there's lots of definitions of a robot. I'll just stay with a simple one. It's a machine that senses, thinks and acts upon the world.

So there's three things right there that we have to deal with: a) Sensing: the ability to know what the temperature in the room is or the ability to know how far it is from a wall. b) It has to be able to think. This is where our virtual intelligence comes in, or other forms of intelligence. It has to be able to make sense out of things it's sensed. Perception. c) And then it has to make certain decisions of what to do and then act upon the world.

The action comes with actuators. This doesn't always mean moving around, but could would be something like starting a siren device. Something has happened, the robot has detected some entities that are dangerous and needs to start informing people. So making a phone call to a modem could be an action that a robot does.

Or it can move around or it can engage into a fight or a battle or whatever. Another way to think of it is as an agent or an intelligent being that is embodied in hardware and situated in the world. That's what makes robotics different from sort of multi-agent systems. Because now we're not agents that live on the internet, we're actually agents that have a physical body and actually are portrayed in the world.

Autonomous robots do things without human intervention. So you can have mobility, be an intelligent agent and still take your controls through a remote control. If you've watched BattleBots, you can see that they're mobile robots, but they are not autonomous because there is someone in the background with a joystick controlling them.

And I'll talk about two areas that are important. Artificial intelligence, which is also a class that we offer and I teach, deals with intelligence, without dealing a lot of embodiment in hardware. If you make your machine play chess, that's artificial intelligence.

Biological inspiration is another important field. More and more people are beginning to realize that a dog or a cat is rather interesting. It can move around, it's smart. How does it do its thing? Can you learn anything from biology that helps us build better machines and better robots?

Robot intelligence is the same as artificial intelligence. Every AI book that you look at has different definitions. The experts don't agree upon everything, but there's certain categories you want to see.

Problem-solving: You would like a robot to be able to solve problems. The question is, should it be a new problem or should it be something it has seen before? If you program a robot to do certain things, what does it do when it finds a situation that's new (i.e., a novel situation)?

Can it reason? Can it reason about the fact that it needs to get power and to get power it needs to plug itself into an outlet.

And then it needs to be able to plan. For a robot to be able to plug into an outlet, it has to find the outlet. So, it's going to have to approach the outlet and then be able to plug in.

And the last one is something that people don't really agree upon. And that is whether you need learning or not. There's a lot of intelligent robots out there that are not capable of learning. And the question remains (and it's still an open question), "Do you need to be able to learn to be intelligent?" We can build you an amazing program that plays backgammon or chess without learning anything from you and still defeat you when playing at the master level. Would you consider that intelligent? Yes. But it doesn't learn.

And there also have been approaches to build a machine that actually starts at almost no knowledge and gets better as time goes on.

History - Electronic Tortoise

This (slide graphic) is probably the first mobile robot ever built. It was built by a British scientist, Sir Grey Walter. In 1950s, he built these electronic tortoises and they were built using vacuum tubes and wheels and plastics and such. And he had lights on them and he had light sensors in them, and he could actually have them interact because each could sense the light of the other one and deal with it.

And he did some cool experiments where he would actually put one of these things in front of a mirror and then it could see itself. So then he was dealing with issues of self-reflection and stuff. And he would publish papers based on this in scientific journals and in places like Life magazine. So he was one of the pioneers of mobile robotics.

History - Hopkin's Beast

Universities got into this business in the '60s. Johns Hopkins University had a strong robotics program at the time and they built something called a Hopkins Beast.

And it was novel for its time because it could actually locate an outlet and plug itself in to recharge. This(pointing to slide graphic) is old technology of course, so the charger here is very heavy but you can see it's trying to recharge. It was nice to see that.

Now look at the image and notice that it's a very, very controlled environment. It's a flat floor. You don't see any carpets. You don't see any obstacles. The height of these things (points to plug) are set,. If you take this robot to a different room, where the outlets are at a different height, it's not going to be able to do it. The colors of this is not too good, but you can actually see that they plug themselves in. This was an initial step to say, "hey, maybe robots can provide themselves with power."

History - Phony Pony

At the University of Southern California (where I studied and got my Ph.D from), they built something called a Phony Pony in 1968. This was a robot that could walk. It was the first walking machine, if you will.

I guess I should tell you that it was controlled by a computer that was the size of a room. Computers were rather large at the time.

One of the things that they did with this robot was make the feet very wide. That's how you can build stability. If you want something stable, either work on the software and control it well or actually make the feet wide so it can't fall.

History - Shakey

Stanford Research Institute, SRI, got into the game and they produced some very interesting robots. Some were very good robots.

This is Shakey with 1970s technology. You can start see that it has a radio link, it has television cameras, it has range finders. It was rather sophisticated.

And this thing would go into different people's offices at Stanford and collect cans of soda. So, it would go in a room and it couldn't find the can, but it would come to you and ask for soda, and then you would give it to him. So he could navigate from one room to another room, which is not bad.

History - MIT

MIT got into the game in the 1980s. A professor named Rodney Brooks sort of started this tradition. and this was actually what interested me in robotics.

And his theory on robotics was that, "let's build them small, cheap and easy to control". Of course, he started with that idea, but this robot, I think, is about 50 thousand dollars if you want to buy it .

This is a six-legged robot and it's pulled using 12 motors, lots of sensors and such and it can, as you can see, actually climb over a book.

The way it can do it is that it raises its leg ... if it clears it, it's gone. If it doesn't clear it, it goes up ...clears and it goes. So it can do a lot of stuff.

And it appears to be a lot more intelligent than it is in reality. And that's something philosophers in AI discuss a lot. When you can observe something producing intelligent behavior, is it truly intelligent or does it just appear to be intelligent through its interactions with the environment?

History - Carnegie Mellon

Carnegie-Mellon University, CMU, took a different approach. They said, we don't want to build something that's not reliable or dependable. Let's make large, extremely dependable machines.

So you can actually stand underneath this robot. It was called Ambler and it was intended for exploration of other planets.

So there are two basic approaches. One says, make robots smaller, make them maybe not as robust, but let's have a lot of them, and they can be cheaper. And the other approach says, "no. let's build very robust, highly redundant machines." They will be larger, they will be heavier and they will be stronger.

And groups with both approaches are competing for Mars exploration. So there were discussions. Of course you have seen it now. But there were discussions that NASA was going to send something to Mars. What would be the right way to move around? So this is sort of the other, the second approach.

Japan's Approach To Robotics

Japan got into the game and I have selected these two for a reason. I spent two years in Japan doing robotics research and looking at it, what is interesting is, it has two approaches for the problem.

One approach says, let's build something that is extremely useful that has lots of applications. This project wasn't completed, but this is a seeing eye robot. So instead of having a cane,the blind will have one of these robots that moves in front of them. It has an array of sensors and it can tell them what's in front of them and what's not. Very, very applied research.

What's nice about Japan is that when the economy was good, they also had this line of research that said, let's do cool things and as a result of doing cool things, cool things will happen and we'll come up with applications. We don't have to have an application line, let's just build interesting things.

This is a robot built at Waseda University in Japan, in Tokyo, and this thing plays piano. You know, if you want to produce music, most of you can do that with twenty dollar software from Office Depot. Right? You buy a synthesizer machine and then you plug in some things and it plays music. This thing actually read musical scores, using a camera. And it had feet that it could press the pedal. And it actually has fingers that could press the keys. Now that's a lot of work to play music. Right? This music score that's here, you could simply feed that to a computer and this synthesizer would produce the music.

So it's hard to justify this multi-million dollar project, but it was very neat, because in the process of doing this, you come across interesting problems. Vision: How do we see and process things? Agility, dexterity: How do you play? ... I mean, I can't play the piano! So those of you who have tried know it's a very difficult thing to do. To train a robot to do that is hard. But here's the nice thing. This robot will play the same piece exactly the same every time, because the timing and everything will not change. So it's important to remember these are the two approaches to doing research.


Then a group of people got together and they said, "Let's look at service industries. We need people to bring us our slippers, and we want robots to bring us our beer and champagne, and we can just sit there and order them around. So a number of companies at the time were formed (this is the early seventies) to do this. And guess what? All went out of business. Because it sounds intriguing and it sounds easy to do that, but do you know how difficult it is to build a robot that moves in my home and your home, and somebody else's home? Some people have carpets, some have multiple floors, some have dogs and cats, some have laundry all over the floor. How do you go around these things? It's a difficult thing. So there were interesting attempts to do this. So, this is ComRoToT from ComRo Incorporated.

Current Status

So currently, where are we? Not much intelligence, unfortunately. We are mostly in structured environments. We like to be in structured environments because it's easy.

There are two things, sort of, that people are playing around with. One is behavior-based control. And, in simple terms, behavior-based control says that don't do so much complicated sensor fusion, but control your ability, your movements, based on the behavior. Now, for instance, I will have a wandering behavior. When I have a wandering behavior, I will just wander. Then, I have a hunger behavior. When it kicks in, I am hungry now, I no longer wander, now I look for the plug. Then I have an avoidance behavior. When I see this I have to avoid it. So you combine these things and have different behaviors subsume each other (called subsumption architecture). and you can get interesting behavior patterns, and this makes programming a lot easier. And of course we'll be discussing all this in the corresponding slides.

Learning has become a big deal in robotic systems. You want to be able to deal with new situations so why not equip your robot with the ability to learn. So some of the approaches that have been taken: trial and error, reinforcement learning. I won't go over these because each one of these is sort of an AI lecture. But I just wanted you to have seen the key words.

Learning by showing: You know there were attempts where you actually stood in front of a robot and did the motion and hoped that the robot learned.

Machine learning has become a huge area. I used to take graduate classes in nothing but machine learning.

And then there are sort of a cluster of approaches called soft computing that have proved beneficial to robotics. and we have borrowed from these fields. I won't go over them, but I jwant you to have seen the key words: genetic algorithms, neural networks and fuzzy logic. These have been very good techniques in AI or in soft computing that have been applied to learning in robotics.


Why do we need or how can we use mobile robots? A big application is health care and elder care. I'll talk about that as I show you some pictures. This is a very common use now.

Agriculture is less prominent. It's a matter of cost.

Let's talk about the top one(health and elder care). Japan is very much pursuing that because they have a problem in Japan. People live too long. To most of you, that's not a problem. But it is a problem because, in the Japanese culture when the people are older and they have to stay at home, children will take care of them. But now people are leaving the work force to stay home and take care of their parents. So that has become a problem. So the government of Japan is investing a lot of money into the building of robots that actually help take care elders.

Agriculture: We do have the technology. There are actually robots out there that can pick watermelons. And there are robots out there that using a visual system and a probe can tell you if a tomato is ripe. But right now it is still a lot cheaper to just use manual labor to do the agriculture. The ability is there, it's the cost. Imagine having these in a farm and someone has to go out there and maintain the robot. That's not a cheap process.

Construction: More and more you are seeing this application because these are large machines that do the large stuff. So that's an up-and-coming one.

Food processing is very big. If a robot was able to make your food you would never find a hair. It's very clean. And also you will always get exactly the same burger. So that's a big field.

Mobility In Unusual Environments: I will talk about that in terms of one of the projects we are working on here which is going to Greenland and Antarctica, and working in harsh environments. So I'll will come back to that.

Personal Service: This is a big deal. Wouldn't you like to have a robot at home that picked up your laundry and helped you brush your teeth?

And the big one is actually entertainment. This is, believe it or not, the largest export of the United States, so robotics should take part in that. And I will talk about pets and amusement parks. When you go through Disney or Universal Studios, and you see those things coming at you, King Kongs and stuff -- robots.

Space: You guys have all seen the Mars Rovers, so I won"t spend too much time talking about that.

Military, of course: There is more and more pressure. I think there is a mandate by the Congress where, by 2013, if I am correct, a percentage of vehicles have to be autonomous. So the military is really pursuing this, because there is no loss of life. They are just machines.

And you see UGVs: The "U" initially meant "unmanned". But of course you can cheat. So now it stands for "Uninhabited" or "Uncrewed". So we have uncrewed or uninhabited aerial vehicles. UAVs are a big deal. And also there is smaller versions of UAVs called Micro Air Verhicles. MAVs.

Personal Service: That's a big deal. Personal robotics are up-and-coming.

Rehabilitation And Feeding: It's a lot easier to depend on a machine and they have done psychological studies of that. If a person helps you in your house, you feel like you are dependent. You don't like it that much. But if a machine's helping you, it's not a problem.

And deliveries of medicine in hospitals - more and more of that is going on.

Vacuuming and lawn mowing, that's a tedious process and it's not that complicated. Although it is, as I will show you and we will talk about that. And gasoline pumping and stuff. The list goes on and on. We'll just briefly talk about that.

Why Can't I Buy A Personal Service Robot?

Why can't you and I go to Wal-Mart and buy a robot that does some of this stuff for us? a) Cost of production! These things are very expensive to make especially at the very beginning. You guys are too young to remember this, but when calculators first came out, they were very expensive, and now, they are a dollar each. So it's the initial cost of production. b) Reliability! It's one thing selling your robot at JC Penneys. It's another thing to fix it. Who's going to fix it? When you buy software and it doesn't work you just call someone on the phone and it works. But who's going to come to your house and take the robot and service it and repair it and stuff? c) Investment capital needs. We won't talk too much about that. d) Liability issues: Now that's a big deal. Unfortunately, before you start doing anything, you have to make sure that you know there aren't a lot of liabilities. And if you have a machine moving around in someone's house, that's a lot of liabilities, right? You could be sued for millions of dollars because you stepped on a cat's toe. So this is something that has to be resolved.

Robots in Space

So let me show you some cool pictures. This is space. This is a very very big field for robotics. It's very important as it's hard to survive out there. This (points to graphic) is a Sojourner which was in '97. Hopefully, you can picture this thing moving around Mars. This was the first Mars rover. This (another graphic) is another attempt.

When they do an extra-vehicular activity, in other words, when they had to be outside, NASA is very interested to have assistance for the humans. But all the tools are built for humans. So then why not have anthropomorphic robots? That's an interesting area of research in robotics, and says, "Look, if you build the robot (this graphic is just an upper-torso robot) so it looks like a human, then it will be a lot easier for a partner to work with it, and then if it can have hands like a human, it can use the same tools we do, and it can hand out stuff and do other things."

So this would be used in a space station or a shuttle, doing outside stuff, that you really don't want to send your astronauts to do.

These are some recent pictures of the Mars rover, Spirit. And you guys have seen pictures of Spirit and Opportunity moving around. Look at the kind of elaborate packing techniques they have to do. We have a much easier time, we just put ours in a crate and send it to Greenland. They have to be a lot more careful. This is the kind of testing they do in places like California.

Autonomous Vehicles

Our Government is very interested in autonomous vehicles. This is a series of vehicles built by Carnegie-Mellon University. And the intent there was to actually navigate across the United States under autonomous control. So these have amazing cameras and image processing techniques where they take pictures of the road, process it and actually steer and drive these things. And they drove that from California to Pittsburgh a few years ago, which was an amazing thing.

One thing that I should also tell you is that 97% of the time, it was under computer control. Three percent of the time it was not.It had to be controlled by humans when it exited or entered the freeways or went down side streets.

See, when you are driving on the freeway you can pay less attention. Though you are going around 70 miles an hour, you won't have cats and bicyclists and stuff running around. But once you are on side streets that's when you have problems. Bicycles actually proved to be a big problem for the vision system, because it's hard to see one. especially if they are sort of directly in front.

Robots for Harsh Environments

This is the project we were working on here at KU, and this is Dante, a Carnegie-Mellon University robot. This robot was actually sent inside a crater, Mt. Erebus, to take some pictures. This is a huge robot. It is lifted using a helicopter and this is a fiber-optic link. This rover was intended to go down into the crater. I think the first day it actually stepped on its own tether which was fiber optic and it was very cold and it was frozen and so it snapped. And they had to bring it out and I think the next day it sort of tumbled. That's a problem. So it had to be air-lifted out.

But it was an early attempt by Carnegie-Mellon to build robots that actually survive in harsh environments.

Here is Nomad. It's another Carnegie-Mellon University robot. It was meant to go into Antarctica and look for meteorites. It was intended to be autonomous. They called it autonomous, but I think every time you see it in motion, there's about five graduate students pampering it and helping it do things. And there's a story, which I don't know if its true or not, but that it could only see meteorites if a graduate student pointed to one with a laser pointer. But it proved that robots can survive and function in these harsh environments which was an amazing deal.


At KU we're working on a robot. This (graphic) is something that my graduate students and I have built. It's part of the PRISM project, the Polar Radar. I guess you've seen talks by PRISM project.

And the idea is to take radars to Greenland and Antarctica and measure characteristics of ice. The thickness of the ice. Where is water in the layers and such? And as part of this project, they need robots to move in a special coordinated manner. And robots should be good at doing precise, coordinated movements. So this is our rover, last year in Greenland doing some movements. At the time it was not fully autonomous. You can see we have a passenger on board. It was an all-terrain vehicle. Tracked. And we outfitted it with actuators, linear actuators, that actually helped it steer and drive.

We use simulation programs to actually test the vehicle before we finalize it. For instance, you can build simulation programs that are true to reality. So this has physics built in. And you could have it actually drag around an antenna. And, you can see how you should drag your antenna. We could test things like: Should it be wheeled or should it be tracked?

There's a lot of simulation programs that are out there. They are more than just animation. You can actually build your robotics platforms and study the kinematics and dynamics of the movement. That helps you with the design.

Here's the robot (graphic) with one of my graduate students, Hans Harmon, and it's actually moving. There is nobody on board.

This thing takes via points. So you give it a GPS coordinate, a sequence of them and it goes. For now, it does it in an obstacle-free environment. But, if you run into an obstacle in polar regions, you have bigger problems than you thought. Because there really shouldn't be an obstacle. So this thing actually gets the GPS location, and it goes where it needs to.

Components of the PRISM Rover

Here's a camera that can take pictures. You can tell right now that it is pointing down at the ground. This is the GPS antenna. And this is the laser range finder that's on board, but it's not functional yet. We're working on the next level which is obstacle avoidance. This is actually a winch that has helped us get out of some bad situations. Here's a picture of the three graduate students.

Flying Robots

You shouldn't just think of aircraft. There's all different approaches to build flying robots.

I spent a summer at the naval research lab working on coordination of these Micro Air Vehicles. The idea is this. If you want to do intelligence gathering, why send one large vehicle where they can just shoot down. Send multiple of these things, they are harder to see, and they are distributed, so they are more robust. And each one of them will take a small picture, and through the powers of software, maybe we can put these pictures together and make a coordinated picture of the world.

Homeland defense

That's another area where using swarm intelligence of group robots is really interesting.

Robot Learning

Let's go back to MIT. I'll show you a picture of an insect, if you will. Next they built Coco which, in some sense, takes the same steps as evolution. This was intended to have the intelligence of a gorilla.

And then, this is the latest. and the most recent project. This is Roger Brooks. this is COG. and let me tell you what's interesting about this robot. The builders of this robot have decided that you shouldn't program your robot. You should give it child-like intelligence. You should give it the ability to learn, and have it learn. And because they wanted it to learn from people, it had to interact with people. But you are a lot more likely to interact with an anthropomorphic robot than a cylinder. So, they built it as the shape of a human, they gave it human ability, so then it plays with people. And the hope is through interactions with people the robot can play with toys like this Slinky and ask you questions about it. This robot is upper torso only - the base is not mobile. They are working on building its legs and stuff. Very complicated for my liking. But for now, it has two arms and an articulated head and it can interact with people.

So this is a project to watch. Because the idea is this. Can we just build simple robots and have them become better as time goes on and as they learn?


You may have seen pictures of this. Honda. I was in Japan when Honda unveiled this and it was interesting because they had never talked about it beforehand. A lot of times when there is work in progress, you hear about it by going to scientific conferences, by reading journals, but Honda actually did this quietly.

So they put, I believe, 10 engineers for 10 years, millions of dollars, to build this. And what was amazing about this, is that they just thought it would be cool. So there was a Honda executive who had a lot of power, and decided, "We are going to build a very nice walking robot." Now you can't do that, right? In many companies, they'd say, "Well, we build cars, why a walking robot?" But they had to deal with a lot of issues like actuation. They had to come up with lot of nice motors to do this. They give it the stability by putting a huge battery, about 100 pounds, in the backpack. So this allows it to sorta stand up.

I've seen this one walk upstairs and walk downstairs. It's an extremely agile robot and it can do that. It's got what's called a dynamic walk.

When an insect walks, it always has three legs on the ground. So it has three legs, three legs, three legs. So at each moment, the center of gravity is within the triangle. So the insect is never falling. But when a human walks, you are actually falling and then you regain, and then you are falling and regain. So that's a dynamic walk. Anyway, this robot could do a dynamic walk.


PRISM © 2002, 2003 is brought to you by
NSF Logo
National Science Foundation University of Kansas
NASA logo
KTEC logo
Kansas Technology Enterprise Corporation
KU Jayhawk logo
University of Kansas