Carnegie Mellon University Prepares for the Self-Driving Future

When many in higher education think about self-driving car technology, we’re tempted to jump straight to robotics, but autonomous car manufacturers are also going to be looking for generalists who understand the bigger picture of autonomous vehicle design and manufacturing.  

How can universities help prepare students to work in the autonomous vehicle industry?

Dr. John Dolan, Principal Systems Scientist at Carnegie Mellon University, joined the Enrollment Growth University  podcast to discuss how higher ed can start preparing for the future of autonomous vehicles.

Higher Ed’s Role in Autonomous Vehicle Design & Manufacturing

“In the master’s program that I direct,” Jon said, “we are trying to give the students an understanding of systems engineering and its business aspects. More broadly, there are policy aspects to consider, which is one of the reasons why intelligent cars or self-driving cars have not been out there sooner.” 

Education could be another reason. Europeans and schools in some other countries such as China and India tend to have many programs in mechatronics, which is a combination of mechanisms and mechanism design. Mechatronics students also learn sensing, electronics, and computer programming and control those three things. 

“We have some programs in the U.S. in that area but not many,” John said. “So that’d be one thing that I think perhaps we’d want to emphasize more.”

We have some robotics degrees in the United States, though. In fact, the earliest PhD program was established at Carnegie Mellon, but that’s also an area in which we need to think about whether we’re going to continue with a divided model, which splits activities among various departments in robotics, including mechanical engineering, electrical engineering, and computer science. The alternative approach would be having a centralized department such as the one at Carnegie Mellon.

What’s Standing in the Way of autonomous Vehicles Right Now?

“There are two general scenarios,” Jon said. “One would be the highway scenarios and then the other would be the urban ones.” 

Under good conditions, we can do highway scenarios now. We have access to the GPS out there without the interference of tall buildings and if it’s a sunny day you don’t have to worry about bad weather. 

Cities, however, pose a lot more complications. It’s difficult even for humans to deal with all the confusion that can go on there. Companies that are working on this problem now going through repeated cycles of hundreds of thousands or even millions of miles of driving in order to try to experience these problems and learn how to deal with them. 

“But I think that’s going to take a long time,” John said. 

When it comes to liability issues, insurance, and things like that, we just haven’t matured sufficiently in those areas to know how to handle them. As we know from the accident that occurred back in March of 2018 with the Uber car, there’s still not a methodology that allows us to rigorously state in a way similar to what is done in the airline industry that cars will be reliable to a certain percentage of safety.

What’s the Long-term Vision for Autonomous Vehicles?

Machine learning and deep learning are more steadily applied now. In the old days, which in this field means four or five years ago, people applied those ideas more or less exclusively to the perception aspect. Now, however, they’ve started to apply it in the lab and to behaviors and predictions. 

“So can you use machine learning to better predict what pedestrians or other cars are going to do?” John asked us. “Can you use it to generate behaviors or to interpret a lot of driving data?”

Maybe! Driving data from the highways in Germany and Los Angeles could help us figure out how humans do lane changes and to learn from that. Another nut waiting to be cracked is the question of how  you verify these systems so that you have assured autonomy or some measure of how safely they’re going to be able to drive. 

“We don’t want to have zillions of cars on the road where we can’t really say very well what they’re going to do,” John said.

Next-Steps for Universities Preparing for a Self-Driving Future

Along with a good foundation in engineering, math, and learning software, consider offering a robotics curriculum. The interdisciplinary nature of such a program is hard to achieve by spreading the same activities around in traditional departments.

“I mentioned systems engineering and project management,” John said. Also machine learning can be used across a wide range of different activities that are associated with autonomous driving.

“I’m involved right now with a DARPA project called assured autonomy,” John told us, “and the question is when you put learning based or learning enabled components into an autonomous system, what impact does that have on the reliability? Can you make some kind of analytical statements about that?”

Answering these complex questions will help schools determine where they can best fit into the landscape of autonomous transportation.

This post is based on a podcast interview with Dr. John Dolan from Carnegie Mellon University. To hear this episode, and many more like it, you can subscribe to Enrollment Growth University.

If you don’t use iTunes, you can listen to every episode here.

Eric Olsen

Eric brings more than a decade of award-winning creative brand development, marketing analytics and higher education experience to Helix Education. Eric is a graduate of Bradley University and earned his MBA at Lewis University.