Challenges in materializing fully automated vehicles
Self-driving vehicles seem to be just around the corner. However, fully automated vehicles still haven’t materialized yet. What are the challenges faced in this? We sat down with Professor Dariu Gavrila, head at the Intelligent Vehicles Group at the Department of Cognitive Robotics of the renowned Technical University Delft in the Netherlands. In this interview he outlines the current status and future prospects of this transformational technology, and especially how implementation of robotics at scale needs to be aligned with actual human behavior and interaction.
Please, can you introduce yourself?
My name is Dariu Gavrila. Over the past twenty years I have been focusing on facial systems for detecting humans and their activity with application to intelligent vehicles, smart surveillance and social robotics. I led the pedestrian detection research at Daimler R&D, which was commercialized in several Mercedes-Benz models. Currently, I am head at the Intelligent Vehicles Group at the Technical University of Delft where I perform research on self-driving cars in complex urban environments.
Why is there potential in self-driving cars?
There are several reasons. 1.35 million people are killed yearly in traffic worldwide and 95% of the accidents are due to human error. If we could automate driving, the hope is that for a very large part, these accidents can be avoided. Also, certain people with disabilities obviously might profit from self-driving vehicles.. On top of that, there is a comfort case. Instead of pedaling and steering, you could actually do something social, nice or productive while in a vehicle. Finally, there is a very strong economic case in the sense that, certainly in the business of transporting people and goods, the hope is that by removing the driver, you can save a lot of costs. That’s why there is potential for this technology to revolutionize the mobility in the future.
How does such a revolutionary potential emerge?
When revolutionary technologies emerge , it’s very instructive to take a look at the so called Gartner hype cycle. The Gartner hype cycle is a development model for new emerging technologies from conception to possible maturity.
It starts with a certain technological trigger. Let me present some proof of concepts in this case of self-driving vehicles: People get excited, the media gets on board, then many startups start appearing, people start increasing their expectations, they come up with even shorter deadlines and announcements. And at some point we are here at the peak of inflated expectations and then suddenly things start to significantly slow down. People get disappointed. The problem turns out to be harder than they thought, which results in a washout effect, where a lot of consolidation takes place in the industry, many players go down and perhaps get taken over by bigger players. After a certain kind of realization of the main problems, at some point the cyrcle goes upwards again as these problems are taken care of, which results in some realistic applications and commercial products appearing. This is the so called slope of enlightenment up to the plateau of productivity.
How does this apply to self-driving cars?
At the peak of inflated expectations, Chris Urmson, the Car Chief at Google, said: “Okay. my team is committed to making sure that my son doesn’t need a driver’s license in five years.” Ford CEO said: “Well, we expect fully autonomous cars in five years”. And of course, Elon Musk topped it all up and said, well: “We’ll get fully autonomous Tesla by 2018”.
We had two years of COVID, so we can add two years. But we’re now in 2022 and… Where are all these cars? Why aren’t these cars self-driving? First of all, there are still some open legal issues, like regarding the revision of laws and harmonizing over multiple countries. There are still rules that need to be worked out for the validation of the technology of how much an OEM needs to test a certain technology before it can be introduced in the marketplace. Of course, there are issues about ethics and liability, but I think these are not the bottlenecks for this technology. We have the human factor issues, and these are actually, very important to see whether this technology will make it into the marketplace. And so, it’s a question about people, whether they accept self-driving technology,, whether they’re actually willing to buy and trust it. So, what is the true bottleneck?
What are the specific bottlenecks in self-driving Technology?
Sensors (cameras, radars and laser scanners) experience problems with darkness in tunnels and heavy rain or snow. If it’s slippery, problems with the vehicle dynamics emerge. Also, some objects can be really hard to detect.
For example, if there’s a green light, it does not necessarily have to be a traffic light and these things can confuse sensors. Certainly, while blinded by the sun it can be hard to see these things like this for humans, but it’s also hard to see for a self-driving vehicle.
Self-driving cars need to have sensors to already detect very small debris, maybe 20, 30 centimeters at distances of 150 meters. There are so many scenarios at play here: urban environments, dark shadows, and object classifications triggered by AI neural networks and motion prediction (pedestrians crossing, cyclists turning). In my presentation at the last DIA Event, I took a deep plunge in some of these scenarios, from the pantheon in Rome to Phoenix in the United States.
Where do we stand?
We can distinguish several levels of automation:
0 – No automation
1 – Driver assistance & Feet off
2 – Hands off
3 – Conditional Automation & Eyes off
4 – Brain off (Robot taxi)
5 – No steering of wheel possible anymore
We now have the first L3 system (where you are legally allowed to do side activities) since December in Europe. In Germany, it’s part of the Mercedes-Benz S-Class and well, what you do is you turn this system on and the car drives itself and it does the longitudinal, the steering and the braking by itself. The car will give the driver ten seconds to take back control if there is a need to. Being allowed to do side activities gains you quality time in the car. However, there are still many limitations to the system: A self-driving vehicle only drives up to 60 kilometers per hour. It cannot overtake. There has to be a car in front. So there are some pesky conditions, but it’s a first start. There are now a couple of routes in San Francisco where people can actually order their cars and drive without a driver. It’s very cool. Apart from driving people around, transporting goods like groceries or packages is going to be a major application of self-driving vehicles, which is very exciting in the logistical sector.
What’s your prediction on the future?
Right now we have started hands off highway pilots and remote parking pilots. As I mentioned before, the Mercedes S-Class is the first one for autopilot with eyes off where you can do side activities. Hence, in the next few years, we will see these systems under many more regular conditions. Meaning that self-driving cars won’t only go up to 60 kilometers, but more like driving 120, 130, overtaking etc.
I think it will take until 2030 where you can really have some type of robot taxi for the highway. We may see some nice installations of automated valet parking. That means that you go to some parking lots and when you step out of the car, it will park itself. Maybe, some fancy restaurants will be offering this. Thus, personally, I think that it will take till after 2040, until these vehicles have the same flexibility as humans, while driving around cities like Amsterdam. So, are self-driving vehicles around the corner? Yes, they are. Some have already arrived. Some are just around the corner, and others will take quite some time to get to the corner.