Patience is pivotal for the autonomous vehicle future
The fatal collision between an Uber ATG vehicle and a pedestrian was a reminder that autonomous vehicles are not ready and that a difficult technological hill remains.
The commercially available Level 5 autonomous vehicle won't be here in the next few years and it may never arrive. The path for a new technology is seldom straightforward, and for autonomous vehicles the path was complicated by the fatal collision between an Uber ATG self-driving vehicle and a pedestrian in 2018.
The National Transportation Safety Board found that the immediate cause of the collision was a failure of the Uber ATG operator to closely monitor the road but also sited Uber ATG's inadequate safety risk assessment procedures. The system, as well as the operator, both failed to react properly to the changing conditions of the road. Regardless of fault, the accident raised levels of mistrust and skepticism in public support for autonomous vehicle research.
Two years later, the timeline for the fully autonomous vehicle future remains blurry despite the major technological advances made during this time. There are numerous vehicles that have automated features and attributes on the road already but establishing how far off a true Level 5 autonomous vehicle is from now is nearly impossible -- even for experts.
Where we are now?
The most important thing for people to remember is there are no autonomous vehicles available for sale, said Sam Abuelsamid, principal research analyst leading Navigant Research's mobility research.
"In fact, there aren't really any truly autonomous vehicles anywhere. Everything that's out there still requires some degree of human supervision," Abuelsamid said.
While there are no fully autonomous vehicles, there are numerous examples of companies employing vehicles of varying levels of autonomy. On top of these several technologies can be sited as important steps toward full autonomy. Braking and steering are both increasingly being automated. Emergency braking at a basic level could be considered the first step toward autonomy, said Mike Ramsey, senior research director and transportation and mobility analyst at Gartner Inc.
Automatic emergency breaking (AEB) has found its way into more and more vehicles through the previous decade and is a step toward an autonomous vehicle future. Ten automakers, including Tesla, Mercedes Benz, Honda and Toyota, reported equipping over half of their vehicles between Sept. 1, 2017, and Aug. 31, 2018, with automatic emergency braking, according to the National Highway Traffic Safety Administration and the Insurance Institute for Highway Safety.
On top of AEB, technology such as lane assist allows the vehicle to essentially 'steer' itself in certain conditions and adaptive cruise control allows the vehicle to adjust its speed.
Closer to the highest level of autonomy than those technologies are Tesla's autopilot, Cadillac's super cruise and Volvo's pilot assist that combine each of the previous features and can work very well in straightforward situations. These all, however, still require the driver to pay attention and be prepared to engage in certain circumstances.
"If you are not paying close attention to what the system is doing, and ready to take over, then you can get into trouble," Abuelsamid said.
There are numerous obstacles, both literally and figuratively, in the path of achieving a fully autonomous vehicle and the process has been necessarily slower than many predictions.
What stands in the way of autonomy?
According to Abuelsamid there are four basic steps to doing an automated driving system.
- Reception: Taking in sensor data and having software algorithms make sense of that data and understand the environment around the vehicle.
- Prediction: Once you've classified surroundings, then you have to predict what they're going to do in the next three to five seconds.
- Path planning: Once you projected where everything's going to go, then you can plan your path through that environment.
- Control: Sending the signals to the actuators, to the steering and braking and to the power train to operate the vehicle.
At each point in the development process there are significant technological challenges that must be surmounted. Of these steps it is reception that has proved to be a significant challenge. Though the introduction of machine learning algorithms has helped in this process, systems are still not 100% reliable. They need to be as accurate as possible because without a perfect understanding where other road users are and what perceived objects are, predicting becomes impossible. On top of this, if the system identifies the object correctly, prediction remains a significant hurdle.
Human drivers can attest to the unpredictability of driving and how crucial being adaptive can be. Unexpected lane changes, unpredictable pedestrians and aggressive drivers are daily occurrences on streets. Figuring out a way in which a machine can be deployed and not taken advantage of is one of the underappreciated difficulties with this technology. When people understand the limitations behind these systems, they may seek to take advantage.
Autonomous vehicles must be able to assess and adapt just as easily as the human brain and they are not at that stage yet. In the fatal collision between an Uber vehicle and a pedestrian, the system sensed there was an object approaching the road and walking across it but was unable to properly predict that the victim was going to walk directly into the path of the vehicle.
Is Level 5 possible?
The society of automotive engineers defines six levels of automation for automakers, suppliers and policymakers to use to classify a system's sophistication. The lowest level 0 has no automation and the highest level 5 is a completely autonomous vehicle.
According to Abuelsamid, Levels 4 and 5 of autonomous vehicles are incredibly similar in that neither require human intervention and vehicles at either level should be able to get themselves to a safe and minimal risk condition without assistance. The staggering difference is that a Level 5 vehicle should be able to operate anywhere a human can. It must assess, predict, plan and execute as a human might in any situation. This is a monumental task that will take a long time to complete.
Sam AbuelsamidPrincipal research analyst, Navigant Research
"Increasingly, the impression I'm getting is that Level 5 may not ever happen," Abuelsamid said. "There's always going to be limits. There's always going to be some restrictions on where these systems can operate."
The technological hill is only part of this though. With time, perhaps these challenges could be tackled, but there must be a significant, financial reason for companies to find the answers. If there is no market, or a tepid market, for a Level 5 autonomous car, then the likelihood of its creation is quite low. Ramsey said it's questionable that there's an actual market demand for autonomy outside of some specific use cases.
Though automakers see consumers asking for more automated features there remains a question as to whether the investment would be worthwhile from an ROI standpoint to attempt a Level 5 vehicle, thus clouding the possibility of an autonomous vehicle future.
"[The trepidation] is starting to make an impression on the automakers and they're pivoting and saying, 'Well, maybe we should be focusing more on safety technology or in-vehicle technology for comfort or enjoyment,'" Ramsey said.
What does the timeline look like?
For the time being, further advancements in this field are set to remain small scale, small deployment. Logistics seems to be a natural steppingstone for autonomous vehicles, however, and deployment in this field could act as a solid proving ground.
"The first stage is more likely in logistics, where you see essentially factory automation, then it spreads to logistics automation," Ramsey said. "Between warehouses, in ports and rail yards, in areas that are controlled where you're moving packages and not necessarily people."
These areas seem to be the safest and most logical steps for the technology, minimalizing risk to people and allowing for improvement to its capabilities.
"Eventually, so the next three to five years, we'll see limited deployment, the latter half of the decade we'll see that ramp up. At the end of the 2020s, we will probably get to several million autonomous vehicles on the road," Abuelsamid said. "And these are predominantly going to be in commercial applications, whether it's ride-handling, robo-taxi type services or goods delivery services rather than being sold to consumers."
The autonomous vehicle timeline is one that has been extended throughout the years and for good reason. There is a lot of promise with this technology but a lot of risk. Setting hard timelines seems fruitless, especially when so much is left unsolved.
"We have a lot farther to go to prove that this technology is safe, but, as an engineer, I'm also reasonably confident that we will get there in time," Abuelsamid said.