“Boss” the autonomous Chevrolet Tahoe developed by General Motors and Carnegie Mellon University that won the 2007 DARPA Urban Challange
Over the course of the past 30 years, I’ve come to know and respect many engineers and noticed an interesting dichotomy among many of those that work in the field of transportation. They can be at once among the most optimistic and pessimistic people I’ve known. Engineering is all about applying science and technology to develop creative solutions to the problems we face on a daily basis.
Throughout most of human history, mobility often has been fraught with peril. Most of that time we have had to move on foot and while humans have some unique physical advantages such opposable thumbs and our ability to verbalize, we are sadly lacking in speed, strength and endurance compared to many other species. Despite that, we have used our ingenuity to develop tools and devices that enable us to get around more effectively.
That said, the machines we’ve created over the past century frequently have only changed the risks we face rather than eliminated them. In the early years of the automotive era, deaths as a result of traffic accidents were shockingly high. However, since 1960 fatality rates in the United States have dropped by nearly 80 percent from 5.06 deaths per 100 hundred million vehicle miles traveled to just 1.13 in 2012. This can be attributed to vehicles with better handling, braking and safer structures, accident avoidance technologies like electronic stability control and changes in human behavior, particularly reducing drunk driving.
Despite that tremendous progress, more than 33,000 people died on American roads in 2014. The goal of engineers is to bring that number down to zero. While we know that goal will never be achieved, there is still room for a lot of improvement.
Throughout my 22-year career in engineering, I and everyone I worked with were almost always confident in our ability to eventually find a solution to every problem. The key word there is eventually and that brings us back to that dichotomy I mentioned in the opening of this piece.
In attacking any engineering problem, there are two ways to approach it. You can try to fix a symptom with a band-aid solution which is often sub-optimal and leads to further problems down the road, or you can deal with the root cause. Ultimately, we usually end up somewhere in between which is where we are now.
The root cause approach requires stripping a problem down to its bare essentials to understand what is happening. In doing this, a solution often becomes apparent, or at least it seems that way. Let’s take for example, the problem of loss of control during braking. The way tires work, they need to be rolling in order to generate the lateral grip required for steering control. If too much brake pressure is applied and the wheels lock up, the driver cannot steer.
The obvious solution is to detect impending wheel lock and reduce the brake pressure automatically. On a section of straight, flat, dry pavement, this is fairly easy to do by using a sensor on each wheel that compares the speeds and detects if one or more are about to lock. However, that particular condition only accounts for a tiny proportion of the cases where wheels lock up. On ice, snow, gravel, banked curves, split conditions or transitions, it becomes increasingly difficult to detect what is happening. The engineers I worked with during my career came up with all kinds of creative ways to detect what was happening and respond accordingly. However, while the fundamental concept of how anti-lock brakes should work was fairly straightforward, creating computer algorithms to deal with all of the other scenarios that could crop up in the real world took many years and in fact is still not complete.
The 1978 Mercedes-Benz S-Class debuted the first electronic ABS from Bosch
Mercedes-Benz introduced the first electronic ABS for cars in 1978, a system invented by Bosch. It’s now 37 years later and traction control, which is basically ABS in reverse to apply the brakes when wheels are spinning up still cannot reliably detect when a vehicle is in fresh, deep snow. In this condition, it’s actually better to let the wheels spin up so the vehicle can get moving. Traction control stops the wheels and the vehicle gets stuck. For this reason, every vehicle built still has a traction control disable switch that allows the driver to manually turn the system off as I had to do during a recent snowstorm so the vehicle can get going.
Good engineers recognize that once they have initially defined a solution to a problem, that is often only the beginning of a very long road to get to something that is actually suitable for use by normal people. They still have to work through all of unknown, unknowns. Thus engineers are typically happy to share their initial solutions in the hope that their peers can help identify all of the other associated problems along with workable solutions.
That’s where we are today with the self-driving car. We have identified the human driver as the root cause of most of the remaining traffic deaths. By replacing the driver with a set of sensors that can see farther and with increased accuracy along with actuators that can respond more quickly, we have the potential to eliminate accidents. The operative word there is potential. At this point in development, we are nowhere near being able to actually realize that potential.
In the world of autonomous vehicles, we are currently at the stage where ABS was in the 1970s. We have the hardware and software to control a vehicle on smooth, dry roads in good weather. Unfortunately, those conditions only account for a fraction of the accidents that happen every year. Many accidents happen when people are driving too fast for less than ideal conditions such as when it’s raining or snowing or dark.
Currently, full autonomous control requires fusing information from multiple sensors including GPS, radar, ultrasonic, LIDAR and cameras in order to create a complete picture of the surrounding area. If one or more of these sensor systems becomes inoperable, autonomous performance is severely degraded and in many cases is actually far worse than a typical human driver. Unfortunately, most of those sensor systems currently cannot operate in those conditions where we need automation the most.
Engineers recognize this and are striving to develop solutions but this is likely to take many more years. That’s just the problems related to the limitations of current hardware. To date, no one has created reliable solutions for how to program ethics into vehicles, how to keep drivers engaged during autonomous operation or how to deal with interactions with non-autonomous vehicles.
In the realm of developing control software for word processing or calendars or running a phone, bugs are annoying. Similar bugs in control software for a vehicle can be life-threatening, reversing the process we’ve already made. Engineers are well aware of this and thus most of them are reluctant to aggressively promote what they’ve done as the cure to all of our mobility ills. It’s usually the marketers and PR staff that want to make their employers seem like technology leaders that are pushing the stories forward even when it is unwarranted.
In time, we will almost certainly reach a stage where we hop into a vehicle, tell it where we want to go and get out at the other end. That day may be decades away for most of us, but I’m confident we will eventually get there. In the meantime, I want engineers those engineers to stay both optimistic and pessimistic enough to make sure we do it, not necessarily perfectly, but at least good enough so that we are making things better rather than worse.
