Look, No Hands
Chuck Thorpe's Navlab Project is dedicated to creating safer highways. In this story, he demonstrates how his prototype van can analyze the road ahead and steer itself along the highway. Chuck answered questions about the Navlab project in an Ask the Scientists panel following the premiere of the Scientific American Frontiers special Robots Alive! Here are viewers' questions and his answers:
When will commercial versions of the Navlab vehicle or its descendants be on the market, in your estimation?
First, a caveat: the timetable to commercial deployment is set by lots of non-technical factors, such as marketing, liability, and cost. So it's hard for a technology person like me to really answer that accurately.
There will be a progression of technology going to the market, rather than the whole package getting built at once. You can already buy in Japan a vehicle with "smart cruise control" that uses a laser to watch the vehicle in front, and slows down your own vehicle if you get too close. In the US, several groups are testing that kind of smart cruise control, plus warning systems that tell you if someone has cut in front of you and you should hit the brakes yourself. So those systems should be on the market in a few years.
The drowsy driver warning systems, that watch the road and try to wake you up if you are drifting off the side of the road, are the next systems that could be marketed. I would expect to see those in 5-10 years.
Other than those, my crystal ball gets pretty fuzzy. Steering assist systems, that help you hold your lane when you are in a crosswind, for instance, could also be marketed in 5-10 years. Full automation, where you let the computer do all the driving, is probably further off. The National Automated Highway Systems Consortium, of which we are a member, is working right now on that question. We're building the technology, which you saw on the show; but also, at the same time, studying people's reactions to the system, and understanding how automated driving will best fit into the highway system.
Can you share any details about how much a project like Navlab costs and how you get funding for such a project?
Our research funding initially came from DARPA, the Advanced Research Projects Agency of the Department of Defense. Now most of our funding comes from the Department of Transportation. In both those cases, the agencies had competitions for who had the best ideas, the best cost, and the best capabilities. We put a lot of work into writing good proposals saying what we would do, and our proposals were selected.
At first, the project was expensive. The first Navlab was the blue Chevrolet van that we used 10 years ago, and that appeared briefly on the show. It cost over a million dollars to build, because we had to build our own experimental computers and sensors. The costs have come down a long ways from then, but they're still to expensive to put on the market. Once we decide the best set of sensors, we can build sensors and controllers for the same kind of price that you now pay for radios, cruise control, air bags, and other electronics and safety features.
Why did you decide to become a designer of systems for cars of the future? And what you like best about your job?
I thought I was going to be a physician, like my Dad who just retired from working in Zaire for 30 years, until one day in Anatomy and Physiology class when I realized I didn't have the patience to memorize all the bits and pieces of the body. At the same time, I was getting interested in computers. So I did my graduate work in Computer Science. I had the good fortune to get involved with mobile robots right when we started the lab here at Carnegie Mellon in 1979. Our first robots ran indoors, and ran very slowly - they would move a half meter, then stop for a minute as they looked at the world and tried to figure out what to do next. From there, we have been making fairly steady progress, both in going from indoors to outdoors, and in moving faster and more safely.
The best part of my job is that I keep learning all kinds of different things, from all kinds of different people. For building robots, I work with mechanical engineers, electrical engineers, and computer scientists. When we started thinking about how to wake up sleepy drivers, we got to work with psychologists. When we started building automated cars, we got to work with automotive engineers and civil engineers. So I'm always finding new things to explore. Plus, it's fun to work on technology that is both interesting in its own right, and will eventually be really useful in making travel safer.
With your new invention will people even need to be in the drivers seat?
At first, the systems will be there to help people drive, and to wake them up if they start to fall asleep and drift off the road. But eventually, I want to be able to point my car towards my vacation destination, then recline the seat or crawl in the back and take a good long nap.
If we were to use these cars for the future, how would we overcome obstacles we saw on the show, like lights that don't work (with a policeman directing traffic)...and, say, stopping in an emergency?
The first of these systems will run on the highways, where there are many fewer things to look for. We have been practicing having our car detect orange traffic construction barrels, decide that something is going on here, and change lanes or stop the car. We can stop in an emergency in several ways: the radar on the front of the car can see unexpected objects (like people) and hit the brakes, or the driver can take over from the automated system and stop the vehicle.
I read in the 4-09-97 USA Today newspaper that there was talk of building a road that had magnets or something similar to control the car. I was wondering if you had heard of this, and if in the near future you were going to join up or even planning on looking at that plan in an attempt to speed up or make the work and cost involved in your project less.
I was excited to see the article in USA Today. In the National Automated Highway Systems Consortium, one of our partners is the University of California at Berkeley. We are primarily doing vision-based work, they are more focused on "magnetic nails" placed down the center of the road. The magnet following system doesn't need painted lines or headlights on the car, and will work even through snow and rain. But it obviously requires putting down the magnets everywhere you want to run. Also, with today's technology, the magnetometers can only see the magnets when the car is in the center of the lane, so it's hard to change lanes. So it makes sense to continue the work on both magnets and vision systems, and we'll decide later whether we should use one or the other of the systems, or both together.
Hi, I am a twelve year old boy and I have thought about cars driven by computers in the future. When I saw the show I could see that these cars could actually appear on the road soon and I was wondering how the roads would be adapted.
That depends on the particular location we're talking about. For some of the safety applications, we want to be able to use our computer vision systems wherever you drive your car, so we can't go around the country and change all the roads. Instead, we have to make the cars smart enough to handle whatever is out there.
At the other extreme, for congested city freeways, we may go as far as building separate lanes for computer-controlled cars. Then we can do many more things to the lanes, such as put sensors in them to detect ice, put sensors over them to look ahead for stalled cars or things that fell onto the road, or put communications systems throughout the roadway so a central traffic management center can talk to all the cars.
For several months now, I have been interested in creating an automated car. The approach that I have been thinking for guidance is based on the GPS system (in fact it was interesting to note in the Robot flyers segment where they were using a GPS based system to provide accuracy to a cm). As my initial foray into this project I am working with high school students to build a computer controlled radio controlled car. Do you have any ideas and suggestions that might help our work on this project?
Right! We have also been playing with GPS. With raw GPS, the accuracy is only a few tens of meters. That's partly because the atmosphere distorts the signal, and partly because the government can deliberately corrupt the unencoded part of the signal that you and I can read. With differential GPS, there is a ground station that knows where it is, so if it is getting a GPS position that is, say, 10m to the north of its known position, it can broadcast a correction to all the other GPS receivers in the area, and they can move their position 10m south. This gives accuracies of a few meters. The next level is called carrier phase GPS. Without going into the details of how it works, it can give accuracies as good as 2 centimeters. The problem with GPS is that it requires visibility to the satellites, so when you go under a bridge, you lose your position measurement. So for a helicopter, or a farm harvester, GPS is a great solution all by itself. For cars, we're experimenting with combining GPS with other sensors, to handle the problems of bridges and tunnels.
For your car, contact any of the GPS manufacturers and see if they will give you access to one of their carrier phase units, then run in an open parking lot where you have a good shot at the sky.
As I watched your show the other day and watched as the car drove itself down the road, the first thing that popped into my mind was: "how will this integrate with the new technology of personal navigation systems for cars in the future?" I read a lot of popular science/mechanics magazines, and have seen little television screens that show where you are at all times. These display monitors will also allow you to plot in a course and in turn, the car will drive you to the specified destination. I was just wondering if that is where you hope to one day apply your new technology. Hence the subject of this e-mail: integration of systems.
Excellent question. We intend to integrate with in-vehicle navigation systems in two ways. First, for fully automated driving, the computer can use the information from the navigation system to give it cues, such as when an intersection is coming up, when the road narrows, and so forth. Second, when we're using the computer systems to help a person drive, we can use the same kinds of displays both for route guidance and for driver safety warnings.
What kind of backup plan do you have in case of hardware failure and is there any redundancy in your systems in case of failure?
For the experimental vehicles, we have several layers of backup. There are software limits on how fast and how far the steering wheel can turn. Below that, there are hardware limits on the same things. Below that, there is a sensor that detects if the vehicle is accelerating too quickly to the side, and cuts power to the steering motor. There is a big red illuminated switch on the dashboard that the safety driver can hit to take control back. And finally, we sized the motor to be weak enough that the safety driver can take control just by grabbing the steering wheel.
When we get ready to go to production vehicles, we can take everything safety critical that the computers are doing, and turn it all into hardware or even a custom VLSI chip. That will make it both more reliable and small enough that we can build more redundant systems.
What is the power of the computer in your systems and what languages do you use to program the system?
Most of the computing is done on a single Pentium PC. All our code is written in C. The radar that we built has a Texas Instruments C40 digital signal processor for the number crunching and an additional laptop PC.
The old blue Navlab I from 10 years ago had 5 Sun workstations and a homemade Warp supercomputer with 10 custom processors. Since then, commercial computers have gotten a lot faster, and we've gotten a lot smarter about how to use them, so now our computers are not very exotic.
I know the technology for computer driven vehicles is still very new but is there anything in the works for installing some sort of road devices the on-board system could "see" better? Some type of reflector or laser built into the lane the computer alone could detect? Also, will several vehicles be able to "talk to each other"? In the program, a car cut right in front of the van. If both were linked maybe some sort of warning could alert one or both drivers that the maneuver is dangerous at a given time. Just a thought. My husband is a truck driver. You have a really good idea and I hope to see it in use soon.
Good points! We're studying all of those ideas. Some of our partners are working on building magnets into the road, or putting metal strips inside the lane markers that a radar could see. We're doing experiments with radar reflectors along the road, or laser reflectors. All of those systems will work; it is just a question of which system works best, and which is least expensive to install and maintain. We like the vision-based system because it only uses the same markings that are already on the road, so we wouldn't have to install anything special just for these cars.
As far as communications between cars, you're right, the more the vehicles talk to each other the better the system can be. If all the cars are computer controlled, and all talk to each other, the traffic can flow very smoothly and safely. But we don't want to force everyone to equip their cars all at once. So we're trying to make cars that can drive safely even if no other cars communicate with them, but that can also find other smart cars and drive more closely and more safely when they're together. Eventually, there could be lanes just devoted to cars that are equipped with the right radios and computers.
Scientific American Frontiers
Fall 1990 to Spring 2000
Sponsored by GTE Corporation,
now a part of Verizon Communications Inc.