Background:
The number one incentive of self-driving technology is reducing accidence and saving life. In 2015, automobile related fatalities were approximately 1.25 million globally. It is equivalent to five fully loaded Boeing 747 crashing every single day. In the U.S., Human error accounts for 94 percent of all traffic accidents. Vehicle crashes continue to be the leading cause of death for children and adults ages four to 34. The traffic accidence data for 2016 are not available yet. But, based on early estimates, traffic fatalities in 2016 are likely even higher. The latest increase of traffic accidence is likely caused by the increases in driver distractions (such as texting on mobile phones), even the safety of the vehicle has been gradually improved. Self-driving technology has the potential to reduce traffic accidence since autonomous vehicles could have super-human perception, faster decision making and execution, won’t be drunk, distracted, or tired.
Self-driving technology also removes the need of human drivers. It frees many commuters from driving and allows them to spend their commute time on something else. It increases everyone’s mobility, especially the people with disabilities and elderly people. On large scale, autonomous vehicles would make the transportation system more efficient, reduce congestion and carbon emissions, reduce the number of cars needed, reduce the need of parking, lead to more efficient city planning. It will completely change the way society thinks about transportation.
We are all very excited by the potential of the autonomous vehicles and have very high expectations for the self-driving technology. In a perfect world, the self-driving technology would be perfectly safe and no crash would ever happen again and the technology would save about 1.25 million people annually. In reality, autonomous vehicles likely will not eliminate all crashes. In the development and improvement of the self-driving technology, some people may be saved by autonomous vehicles and some people may die or get injured because of using autonomous vehicles. Even if the technology saves lives overall, it may increase safety for some while reducing it for others, since autonomous vehicles present both potential benefits and potential risks. Instead of thinking self-driving technology as the magic bullet that can solve all problems in transportation, we should have modest expectations on its performance and impact.
The self-driving technology has already been used today on public roads even it is far from perfect. Accidences have happened involving autonomous vehicles. For example, Google’s autonomous vehicles have been driving in autonomous mode on public road for approximately 1.3 million miles and were involved in 11 crashes from 2009 to 2015. In May 2015, a Tesla vehicle equipped with Autopilot was involved in a fatal crash.
It is impossible to require the self-driving technology to be perfect before it is tested on public roads and in real driving situations. With the current data-driven approach or “fleet learning” approach, the self-driving technology can only get better with realistic driving data. So it is important to allow autonomous vehicles driving on public roads for testing and data collection while ensuring public safety.
On the other hand, we shouldn’t allow autonomous vehicles on the road without regulations if there are widespread safety concerns when the technology is not matured. If the technology is demonstrated unsafe on public road, the public will lose trust and even gain fear of the technology. The public may even reject the autonomous vehicles completely and there would be little market for the autonomous vehicles. Then adopting autonomous vehicles would be a very slow and long process.
The key is to balance supporting innovation and ensuring safety.
Current status:
The Google self-driving project Waymo has 23 self-driving cars on the public road. The Waymo cars drive under 25 mph (40 km/h) and have trained safety drivers aboard to take over the control when necessary. Between 2009 and now (February 2017), Google had tested their fleet of autonomous vehicles in autonomous mode for a total of 2 million miles (3 million km). The test roads were preselected and the tests were in good weather conditions (no heavy rain or snow). Google release monthly reports on any traffic incidents that involve their self-driving cars. The the California DMV also requires Google to report the number of incidents during testing where the human driver took control. So far, all test cars have been involved in 14 collisions and other drivers were at fault in 13 of the 14 collisions. Only one collision caused minor body injury — three Google employees suffered minor injuries in one incidence in which the other driver was at fault. Google self-driving project carried out their test with great consideration of public safety.
In addition to Google, many companies have started testing and data collection on public roads. In June 2016, GM began testing self-driving Chevrolet Bolt EVs on public roads. Today, GM has 50 self-driving vehicles on the road. Similar to Google, GM autonomous vehicles also have human drivers behind the wheels to monitor the driving.
The first batch of autonomous vehicles in Volvo’s Drive Me project have already hit the roads of Gothenburg in 2017. They also require human driver supervision. Before that, in August 2016, Volvo and Uber formed a partnership in developing self-driving technology. Uber added its own self-developed autonomous driving systems to the Volvo XC90. To move forward and solving insurance issues, in 2015 Volvo announced that Volvo Cars will assume liability for SAE level 4 vehicles if a crash or incident is a result of a defect in the AD technology.
Toyota Chauffeur project has been testing autonomous vehicles on closed courses. Toyota has been focusing more on another project called Guardian that develops next-generation driver-assist safety system. Both projects are recently started in the Toyota Research Institute at the U.S.
Tesla have auto-pilot as optional features on their Models S and X vehicles. As of October 2016, all new Tesla vehicles are equipped with hardware and sensors to allow for fully autonomous driving. However, the full self-driving feature is not enabled and instead operate in a so-called shadow mode. In shadow mode, the sensors on Tesla cars are active and collect data and the autonomous system generates driving control signals, but does not send the signals to actuators to execute. The human driver maintains full-control of the vehicle without even aware of the simulated driving by the autonomous system. Both the sensor data and the simulated control signal are collected and used by Tesla for self-driving development and testing.
Projections:
Almost all autonomous vehicle makers set their level 4 commercially available timeline target around 2021 or 2022. These cars will be able to operate autonomously under normal traffic conditions on selected roads (with a low speed limit). For customers who purchase autonomous vehicles, they would be able to turn on the self-driving feature on some segments of their driving journey. Different from the Tesla autopilot feature, the consumer driver in a level 4 autonomous vehicle doesn’t not need to monitor the driving during those segments. Those level 4 consumer autonomous vehicles would still have steering wheels, gas pedals, and breaks, since the consumer driver would still need to drive the car during bad weather conditions and on other “unselected” roads. Other dedicated autonomous vehicles that only operated on those selected roads would not need to have steering wheel, gas pedal, or break.
The level 4 self-driving feature would first be available on luxury electric or hybrid cars. The add-on self-driving equipment cost and the better safety features in low-speed crashing would requires the autonomous vehicle be significantly more expensive than a standard car. Luxury car owners are usually less sensitive to the vehicle’s cost or other added cost. It would take another 10 years or longer to self-driving as standard features for all vehicles. Car makers like Tesla, GM, Toyota, BMW, Volvo and others would equip self-driving feature on their luxury models. Google probably would license out their self-driving technology and partner with auto makers.
Level 4 self-driving features would allow Uber and Lyft to significantly expand their business. They will either purchase cars equipped with level 4 self-driving features or partner with car makers to make specify autonomous vehicles. Each would have a fleet of autonomous vehicles that can self-drive without any driver in them between the customer’s locations and local parking lots through selected roads. When I am ready to go to work in the morning, I can open my Uber app, select car pool and call a self-driving Uber car to pick me up. I won’t be surprised by seeing that the Uber car driving into my drive-way has nobody in it. Car sharing companies such as Zipcar would be able to do the same thing. Instead of going to a Zipcar parking location, a Zipcar could self-drive and pick me up at a local location. Rental car company would join the competition as well. Instead of going to a rental facility, I can rent a car through my cellular phone and have the car pick me up. The on-demand car sharing would reduce car ownership gradually. But it wouldn’t be necessary to reduce the congestion or the number of cars on the road significantly, since we will see many empty cars self-driving on the road to pick up customers. Certainly less Uber/Lyft drivers would be needed. But temporary drivers would be needed for days with bad weathers.
With level 4 self-driving features, the cost of using Uber/Lyft could be reduced since it removes the driver’s compensation. Now, my Uber fair per day is $16 for a round-trip car-pool between my home and my work. It would around $6,000 a year. If I don’t use Uber, but lease a compact car, it would cost me roughly $2,000 for leasing fee, $1,000 for insurance, and $1,000 gas. So, the cost of owning a car for my daily commute would be between $3,000 and $5,000. It is cheaper and more convenient to own a car than using Uber every day. But if Uber could cut the fair half with self-driving technology, my cost of using Uber every day would reduce to $3,000. It is on-par or cheaper than owning a car. I would be more willing giving up my car.
At the same time, self-driving low-speed package delivery would be feasible with the level 4 self-driving technology. However, it would need to combine with other robotics technology to allow the packages be transported from the vehicle to the front door.
So far, no company has release any projection on level 5 vehicles. The common consent is that it would take 10 or 15 years to achieve level 5 self-driving. Levels 2 and 3 were popular a while ago. Currently they are not the favorites since level 2 or 3 doesn’t provide much benefits to the drivers. Level 3 is less safe since it usually takes a few second for the human driver to understand the situation and take over the control of the vehicle.
It is difficult to project the landscape of level 5 technology. Since the current technology approaches are heavily based on data, car makers would have competing edge over pure technology companies. After deploying level 4 self-driving vehicles, car makers could collect massive data through on-board sensors during autonomous driving and human driving. The data collected would be essential for developing level 5 self-driving technology, since the data from an enormous number of users in very diverse situation would include sensor data in bad weather conditions and unsafe road. Technology companies would either become a car maker or partner with a car maker to obtain such important data. The technology companies could also require new laws on releasing sensor data in all traffic accidences and obtain the needed data this way. A broad data sharing would be difficult since data is essential to competition.
Problems to be solved:
To achieve the exciting 2021 or 2022 timeline for self-driving technology, a few safety questions should be answered.
1. How safe is safe enough.
Intuitively if the autonomous vehicles result less traffic crashes and especially fatal crashes than the average human driver, more lives are saved than lost. Then the autonomous vehicles should be allowed on the road. However, since the self-driving technology really provides a transportation service, it should be compared to limo or taxi services. Therefore, the autonomous vehicles should result less traffic crashes and especially fatal crashes than the average dedicated drivers of licensed limo services. Limo drivers are usually much better than the average driver since limousine companies usually only hire experienced drivers with clean driving record.
2. How to prove the self-driving technology is safe in limited time.
Many companies are performing testing on public roads. However, it would take many years to collect enough data to prove a technology is statistically safe. Many companies deploy a fleet of vehicles that are equipped with the same self-driving technology to speed up the testing. More and more companies use data collected from sensors on regular vehicles to do testing in simulation. However, it is unclear if the simulation results can be transferred into real environments.
The testing may need to be done by third parties. After recent emissions scandals involved Volkswagen, Mercedes-Benz, Honda, Mazda and Mitsubishi, it is unclear if the government can trust the data proved by the auto makers. If we rely on self-reporting by the auto makers, the government would need to set regulations and guidelines on collecting road test data and record keeping.
3. How to remove safety concerns in testing on public road.
The standard way of testing autonomous vehicles is having a human driver behind the wheel and monitoring the self-driving. It doesn’t pose much of body injury risk when they are on road with low speed limits. It would pose dramatically larger safety concerns when the autonomous vehicles are tested on high-speed, complicated road conditions, and in bad weathers (for level 5). Innovations on more advanced safety features would be needed to ensuring the safety of the human driver. It would also encourage the early adoption of the self-driving technology. Even the development of safety features would lessen the marginal benefit of full autonomy, it would eventually help the adoption of the technology by removing fearing of safety from consumers.
Road testing may be carried out in a phase-by-phase manner. The road test in more difficult situations would only be performed after the technology has been proved safe in easier conditions.
Conclusion:
1. Autonomous vehicle should be as safe as limo services.
2. Safety of self-driving technology may need to be evaluated by third parties.
3. Data share would be difficult since data is essential to the development and competition of the technology. Publically available autonomous vehicles sensor data should be encouraged.
4. Auto makers have the competition edge on data collection for the development of level 5 self-driving technology
5. Safety features and technology would be developed in parallel with self-driving technology.
6. Self-driving technology will start to have impact on consumer luxurious vehicles and car sharing industry in 5 years. Its social impact would start to show in 7 to 10 years.