Selfishly, I would like to live as long as possible. The top causes of death among adults in the United States are heart disease, cancer, and automobile accidents. So I attempt to maintain a reasonably healthy diet, exercise regularly, and avoid smoking harmful substances. And I try to both limit my driving and drive in the safest available vehicles.
Living in LA for the last six months, daily long-distance driving has become unavoidable. Consequently, I have continued my studies of the latest in car safety, which, these days, primarily revolves around vehicle autonomy systems. Everything commercially available to consumers today is at SAE autonomy classification level 2 or lower, which means that the human driver must always remain in control and autonomy can only supplement or enhance human driving. Anything above SAE II (essentially anything where the driver is not in control at all times) is subject to high levels of government scrutiny and a very murky regulatory environment.
Within the SAE II and below offerings, there are a variety of systems available, each of which typically operates independently, utilizes one or two types of sensors, and focus on preventing or mitigating one specific type of collision. Examples include lane departure warnings (using the front-facing camera), blind spot vehicle detection (using ultrasonic short-range sensors), and adaptive cruise control (using a combination of front-facing radar and front-facing camera) to maintain a set distance behind the vehicle ahead.
Other fun recent additions include 360 parking assistance (stitching together views from multiple vehicle-mounted cameras with fisheye lenses to show a full view of the car’s surroundings), rear cross-traffic alerting (using ultrasonic sensors to indicate a car approaching in a parking lot), and automatic emergency braking (using camera and radar to detect and attempt to mitigate a low-speed front collision).
Finally we are starting to see more of these systems work in concert, which is beginning to look like the kind of autonomy we have been promised for years. Recent examples combine adaptive cruise control with automatic lane keeping to allow for a reasonable facsimile of autonomous driveway at high speeds on well-marked highways. Tesla’s Autopilot refers to a variety of features, but this particular combination is what most people equate it with.
I recently had a chance to spend several hundred miles testing the ACC and lanekeeping functionality of a 2017 model year Honda Ridgeline. The system was very unpleasant. Lane keeping was inconsistent and could not be relied upon. The adaptive cruise control acceleration and braking was abrupt and uncomfortable. I was using the system in Massachusetts, which means many poorly maintained roads with faded or non-existing lane markings, as well as frequent inclement weather including rain and snow. This resulted in a system that would simply not work much of the time, and was frequently unreliable when it did.
In California, I have tested a 2017 model year Toyota Prius with ACC engaged for several thousand miles. This vehicle does not have lane keeping, and does have the advantage of being used almost exclusively in good weather on wall-marked highways. The system works far better than the Ridgeline — acceleration and breaking is generally smooth and gradual, vehicle spacing is well-maintained, with the car automatically reducing distance in slower stop-and-go traffic and increasing following distance at higher speeds. However, the system is extremely poor at dealing with other vehicles changing lanes.
If a car merges ahead of me, the Prius is slow to see it and often must abruptly brake to maintain distance. On multiple occasions I avoided rear-ending a merging vehicle only by manually intervening. In other cases, the front-facing radar loses its lock on the vehicle ahead, which can cause the system to disengage and requires me to quickly retake control. A few times it has randomly lost its vehicle lock and begun rapidly accelerating, thinking the roadway is clear. Again, I have had to quickly disengage the system by braking in order to avoid a collision.
The recognized leader in vehicle autonomy at present is Tesla, but I don’t have access to test the Autopilot system. Many claim that Volvo’s Pilot Assist is next in line, so I recently test-drove a 2018 model year Volvo XC60 with the full suite of driver assistance and safety systems.
The automated parking was an interesting experience, as was the 360 camera, although I’m not sure they would be useful to me very often. I luckily and unsurprisingly did not have a chance to test features such as collision mitigation braking and pedestrian detection. But I did get a quick look at the ACC + lane keeping systems, which are state-of-the art. With the salesman at the wheel, we cruised down a city road at 35 mph. The vehicle took care of acceleration and braking as well as keeping us in our lane. It also automatically detected speed limit signs and queued up speed changes for the driver to approve, in line with SAE II requirements. To demonstrate the sophistication, the salesman took his hands off the wheel and his feet of the pedals, noting that the system would not stop at red lights automatically unless there was a car ahead — an important caveat.
Everything was going great in the brief demo until the car ahead of us switched lanes and the radar (I assume) failed to achieve a lock on the vehicle proceeding it. Just as in my Prius experience, the car began to rapidly accelerate, even as the vehicle ahead was coming to a stop at a red light. The salesman, confident in the autonomy systems, let the car do its thing for a few beats longer than he should have. At the last second, he slammed the brakes and we avoided an unfortunate accident.
I’m very impressed with the Volvo, but under the hood it is no doubt using some of the same systems, processors, and sensors as every other major vehicle manufacturer. Between the Ridgeline’s inability to work in snow, rain, and on unmarked roads, the Prius’s unfortunate tendency to want to rear-end merging vehicles, and the XC60’s dangerous hiccup during a demo, my conclusion is that we still have a long way to go before we can truly rely upon our cars to keep us safe.
The pace of innovation in this space is pretty speedy given how slowly the industry has traditionally developed. Each of the autonomy systems presents very limited and specific claims. No vehicle maker is shipping a system in which all of the various sensors and systems work together in concert, and there are no commercially available vehicles today that meet SAE autonomy level 3 or greater.
When these safety features work, they are great: a family member’s recent near fender-bender became a fender-scratcher due to automatic braking. And certainly anything that will prevent me from running into a moose in the Vermont woods, or a bicyclist in Pasadena, is a welcome improvement. But at the end of the day, the driver is still fully responsible.
It makes sense that each system has a very narrow operating range, and the default is to not trigger unless the confidence level is extremely high. After all, even a false-positive rate of 1 in 100 is enough to cause people to disable the systems as annoyances or, worse, dangerous. Still, I am disappointed at how far we still need to go, and how far off the dream of truly reliable autonomous “robocars” still seems to be.