The question I am researching is “With specific reference to self-driving cars, what is AI and machine learning and what is the biggest impact they will have on the automobile industry in coming years?” The aim of my project is to find out how AI is implemented into the process of self-driving, and as a whole what is the biggest impact it will have and what are the main considerations that car manufacturers must have when using this new technology. I will structure my essay by first researching briefly what AI and machine learning are and then link it to self-driving cars to have a more appropriate definition for this project. When I go onto looking at the biggest impacts AI will have, I will actually be researching the major drawbacks and considerations that will create obstacles for the development of self-driving cars such as machine morality as well as the enormous opportunities it can bring. Autonomous driving is becoming more and more feasible as companies are continuously announcing their commitment to developing and launching autonomous vehicles. Commercial distribution of fully self-driving cars would see decreased congestion, reduced emissions, more efficient parking, lower transportation costs for everyone as well as a reduction in the cost of new roads and infrastructure. It would also greatly improve the mobility of elderly and disabled people.
The English Oxford Living Dictionary gives this definition for Artificial Intelligence “The theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.” Although there are several variations of this definition, this one is the most suitable for my research. Secondly the definition for machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it learn for themselves. Essentially, the AI in a fully autonomous car will allow it to drive in all situations whether brand new or repeated, without any human assistance. It will also allow the car to “react” to unexpected situations and make a decision to make a certain action. Machine learning in a car is where it will “realise” that it has not performed optimally for a particular situation, so the next time it repeats this, it will adapt the way it operates to be more or fully optimal without it being externally programmed to improve its action by a human.
How does AI and machine learning link to self-driving cars and how does it work?
Artificial intelligence is essential for self-driving cars to function and become a reality within the future. It is needed to have autonomous vehicles on the road safely and effectively. Self-driving cars with AI implemented will be able to perform functions that a human usually would in a safer and hopefully more efficiently, these tasks will involve changing radio volume using voice commands, swerving automatically from dangerous situations and navigating itself using the fastest route being told only a destination. The self-driving cars combine sensors and software to control, navigate, and drive the vehicle. Using AI, most self-driving systems create an internal map of their surroundings, based on a wide array of sensors, like radar. Pioneers in self driving technology like Google and Uber have used lasers, sonar and high-powered cameras along with other sensors, to construct their internal map to identify surrounding objects. AI software then processes those inputs, plots a path, and sends instructions to the vehicle’s “actuators,” which control acceleration, braking, and steering. Predictive modelling, and “smart” object discrimination (i.e., knowing the difference between a bike and a motorcycle) help the software obey traffic rules and navigate obstacles. Partially-autonomous vehicles may require a human driver to take control if the system encounters uncertainty; fully-autonomous vehicles may not even offer a steering wheel. Machine learning is also extremely important as it will allow the car to perform optimally in new situations which is hugely important due to the unpredictable nature of driving. It means that the car can drive anywhere and be able to do it safely and adapt to the new roads without being explicitly programmed. Fully functional machine learning will be extremely difficult to achieve so there are varying levels of automation available in cars. Currently, there are no legally operating, fully-autonomous vehicles in the United States. There are, however, partially-autonomous vehicles—cars and trucks with varying amounts of self-automation, from conventional cars with brake and lane assistance to highly-independent, self-driving prototypes.
There are 5 key levels of automation in cars, each using different levels of AI:
Level 0 -No automation, At Level 0 Autonomy, the driver performs all operating tasks like steering, braking, speeding up or decelerating, etc
Level 1 – Driver assistance, at this level, the vehicle can assist with some functions, but the driver still handles all accelerating, braking, and monitoring of the surrounding environment. An example of this is a car that applies a little extra braking force when approaching another car.
Level 2 – Partial automation, most automakers are currently developing vehicles at this level, where the vehicle can assist with steering or acceleration functions and allow the driver to disengage from some of their tasks. The driver must always be ready to take control of the vehicle and it still responsible for most safety-critical functions and all monitoring of the environment.
Level 3 – Conditional automation, the vehicle itself controls all monitoring of the environment (using sensors like LiDAR). The driver’s attention is still critical at this level but can disengage from “safety critical” functions like braking and leave it to the technology when conditions are safe. Many current Level 3 vehicles require no human attention to the road at speeds under 37 miles per hour.
Level 4 – High automation, the vehicle is capable of steering, braking, accelerating, monitoring the vehicle and roadway as well as responding to events, determining when to change lanes, turn, and use signals. At Level 4, the autonomous driving system would first notify the driver when conditions are safe, and only then does the driver switch the vehicle into this mode. It cannot determine between more dynamic driving situations like traffic jams or a merge onto the highway.
Level 5 – Complete Automation, this level of autonomous driving requires absolutely no human attention. There is no need for pedals, brakes, or a steering wheel, as the autonomous vehicle system controls all critical tasks, monitoring of the environment and identification of unique driving conditions like traffic jams.
These 5 levels of automation are globally recognised amongst all major car brands, currently Based on automaker and technology company estimates, level 4 self-driving cars could be for sale in the next several years but level 5 is still very far into the future as AI and machine learning systems are nowhere near sophisticated enough to enable full automation.
The use of AI in the car industry will be revolutionary and cause serious disruption, which will no doubt have a huge amount of impacts and changes to the industry as well as society as a whole both positive and negative. And despite all the obvious reasons to advance and invest in self driving cars, the use of AI and a computer to drive a car present huge problems and considerations which are currently holding back the development of level 5 autonomous cars. The costs and benefits of self-driving cars are still largely hypothetical. More information is needed to fully assess how they’ll impact drivers, the economy, equity, and environmental and public health.
1. Ethics and machine morality
If motor vehicles are to be fully autonomous and able to function safely and responsibly on roads, they will need to be able to mimic the human decision-making process. But some decisions require more than just obeying of traffic laws and plotting a safe path. For example, which, if any, animals are ok to hit to protect those in the car. It may be safer to continue ahead and strike a squirrel, for instance, than to violently swerve around it and risk losing control of the car. However, larger animals, such as deer and cows, are more likely to cause serious damage to the car and injuries to occupants than a spun-out car. They require a sense of ethics, and this is a notoriously difficult feature to reduce into algorithms for a computer to follow.
With that said, a common scenario that car companies think about is one where the car experiences mechanical failure and is unable to stop. If the car continues in a straight line, it will crash into a number of pedestrians crossing the street. The car is also able to swerve, hitting one person walking on the pavement, and thirdly, the car could swerve and hit a wall, killing the driver. Though this scenario is extremely unlikely to occur, it raised an important question to AI programmers, what should the car do and who should decide? By eliminating human error, it is necessary for the computer to be programmed to make a decision in this situation. This situation was inspired by the trolley problem which was invented by philosophers many years ago to think about ethics and morals. Simply put, there are two options in this situation. Philosopher Jeremy Bentham says that the car should follow utilitarian ethics, it should take the action that will minimize total harm, even if that action will kill the one bystander or even the driver. On the other hand, Immanuel Kant says the car should be programmed to obey duty-bound principles, so that car should not take an action that explicitly harms a human being, and you should let the car take its intended course, even if it’s going to harm more people. What people have concluded is that when using self-driving technology, a series of calculations will be made for example the probability of hitting a certain group of people by swerving one direction versus another and this will lead to a particular action by the car, however no matter the action, there will always be trade-offs. The programming will always involve trade-offs, and these trade-offs will need to be weighed and assessed by the program. But fundamentally the car must be manually programmed to have certain morals and be able to assess social dilemmas. This is where machine morality comes in. This realization came from this “swerve or stay” scenario and prompted an extremely important problem of how to get society to agree on and enforce the trade-offs they’re comfortable with. The first set of regulations announced by the department of transport was for all carmakers to provide ethical consideration and how they were going to deal with it.
Another interesting ethical issue describes an autonomous car about to be involved in an accident, but it is able to select one of two targets in adjacent lanes to swerve into: either a motorcyclist who is wearing a helmet, or a motorcyclist who is not. We presume that neither target will cause any greater threat to the passengers of the autonomous car if hit. The motorcyclist not wearing a helmet is far less likely to survive this collision so it seems reasonable to program the car to hit the motorcyclist wearing a helmet. However, I argue that justice is poorly served here as motorcyclists who wear helmets are essentially being penalized and discriminated against for their responsible decision to wear a helmet. If this were the case, this may even encourage some motorcyclists to refrain from wearing helmets, in order to avoid targeting by autonomous cars. Others will argue that the motorcyclist not wearing the helmet ought to be targeted because he has acted recklessly and therefore is more “deserving” of harm. The point here is as we explore crash-optimisation decisions, we will always end up with social dilemmas and ethical considerations.
In the end, it is very unlikely that people will purchase and use self-driving vehicles unless the car prioritizes their own safety. If the autonomous car was programmed to protect its own occupants, then it would make sense to choose a collision with the lightest object possible. If the choice were between two vehicles, then the car should be programmed to prefer striking a lighter vehicle like a motorbike than a heavier one such as a truck. The most intelligent cars will use AI to be connected to other cars and be able to identify and gather data on the vehicles around it such as the vehicle’s specific model. Using this, the car could automatically choose to collide with a safer vehicle (such as a Volvo SUV that has a reputation for safety) over a car not known for crash-safety (such as a Ford Pinto that’s prone to exploding upon impact). This method may be both legally and ethically better than the previous one of jealously protecting the car’s own occupants. It could minimize lawsuits, because any injury to others would be less severe. But then this may also create an issue as sales may decline for automotive brands known for safety, such as Volvo and Mercedes Benz, as customers attempt to avoid being the preferred targets of crash- optimization systems.
The ethical point here, however, is that no matter which strategy is adopted, programming a car to choose a collision with any particular kind of object over another will require a form of targeting algorithm. These algorithms will end up making larger vehicles statistically safer as they will be involved in less accidents. The owners or operators of these targeted vehicles bear this burden through no fault of their own, other than perhaps that they care about safety or need an SUV to transport a large family.
Many people might disagree and disregard right away that the dilemma above will never occur with driverless cars. It may be suggested that future cars won’t need to confront hard ethical choices, that simply stopping the car or handing control back to the driver is the easy path around ethics. However, I disagree as my research suggests that braking and relinquishing control will not always be enough. Those solutions may be the best solution with current development of technology currently, but if driverless cars are to ever operate outside of limited highway environments, they will need more varied and responsive options.
2. Safety.
Safety is an overarching concern, there are 1.2 million deaths each year from traffic accidents, driverless technology promises to eliminate 90% of these deaths, by eliminating the main source of accidents, human error. 94% of motor vehicle crashes are caused at least in part by human error. The two recent fatal accidents using self-driving cars by Tesla and Uber have resulted in two immediate questions that need answering. Why did the vehicle and its software fail to detect and avoid the pedestrian? And when it did fail, why did the human behind the wheel, not intervene in time to prevent the tragedy? Uber have since suspended all testing of its driverless vehicles while investigations into the cause of the accident are underway. Despite this, self-driving cars don't actually pose any more of a risk to pedestrians, bicyclists or other cars than cars driven by humans. Self-driving cars have already been put through millions of miles of road tests, and experts say that the technology clearly has the potential to be safer than human drivers. Indeed, it could save a huge number of lives if or when it becomes widely adopted. Currently, research suggests that in difficult conditions, such as limited visibility or on bad roads, an experienced human driver may have an edge over self-driving cars using today's technology. But he expects that self-driving cars will eventually be much safer. A key factor to back up this statement is that a self-driving car will have a 360 awareness of its surroundings compared to the far smaller visible range of a human. The self-driving car will also be able to take in far more information than a human driver and cannot get distracted. Self-driving cars are also actually better than humans at obeying traffic laws. They do not drive too fast, they do not text and drive and lastly, they will drastically reduce and hopefully eventually eliminate drunk driving. Researchers say that this alone makes them safer than human operated cars.
The tech companies and carmakers investing billions in developing automated computer systems to replace human drivers will have hoped their technology would be further along
the line before these accidents occurred. However, by studying what happened, developers can optimise the technology used to avoid pedestrians. Self-driving cars are very proficient at detecting and avoiding obstacles. But teaching a car to anticipate that a pedestrian might step in front of it is still one of the most challenging things to program, according to the experts.
In relation to the final point made in the ethics section that suggests stopping and giving back control is the solution to ethics, I argue that this is simply not safe. Many common scenarios exist today in which braking is not the best or safest option, whether by human or self-driving car. A wet road or a tailgater, for example, may make it dangerous to slam the brakes, compared to some other action such as steering around the obstacle or simply through it, if it is a small object. Today, the most advanced self-driving cars cannot detect small objects such as squirrels, therefore, they probably cannot also detect squirrel-sized rocks, potholes, various small animals, and other small but consequential hazards can cause equipment failure, such as tire a punctured tire or sensor errors, or deviations from a safe path. In cases much like this, there may not be enough time to hand control back to the driver. Some simulation experiments suggest that human drivers need up to 40 seconds to regain situation awareness, depending on the distracting activity, e. g., reading or napping – far longer than the 1–2 seconds of reaction time required for typical accident scenarios. This means that the car must be responsible for making decisions when it is difficult to hand control back to the human and expect a responsive and safe action in that time, and again braking might not be the safest action.
Safety therefore will obviously create huge impacts on the automobile industry as AI and machine learning is used.
3. Data
As previously mentioned, self-driving cars create internal maps using their sensors, which generates huge amount of data. To quantify this statement, the average 8-hour transatlantic flight generates 2.7 terabytes of data from sensors and radars. In comparison, just one autonomous car will generate and consume around 40 terabytes of data. This is because cars operate in far more complicated environments compared to an airbus, constantly encountering obstacles and changes to its surroundings. This presents the challenge of how to make use of all this data. Pinpointing data which is of interest is extremely hard. In addition to this, a problem that arises is that when the car has so much information of its users e.g. regular passengers, where they live/work etc. This information could be hacked and seen. Therefore, every aspect of your driving as well as personal information is potentially at risk. Car companies need to think about how to keep particular information secure like they have never had to do before. Jobs like computer science and data scientist will become far more necessary and required by these companies. Leading to research on the moral implications of data security.
Despite these dangers, this gathering of data will create countless benefits and hugely disrupt the way current insurance companies and legal issues are dealt with. For example, in an accident, the police and insurance companies will be able to access huge amounts of information and what happened throughout the accident. For example, if someone was driving the car, what they did leading up to the accident, if they were on their phone, if seatbelts were all fastened etc. This information will be incredibly useful and reliable to see if the driver took an action, if any. And therefore, where to place blame and accountability. The data could also be used to help governments maintain roads and facilities more easily and efficiently. Intelligent cars would be able to automatically detect potholes and other road problems and send this information to the council or relevant group. To inform the car properly and make the very best option in critical danger situations the car will need to analyze every bit of information on its surroundings. Here are a few examples to show how detailed this data can and should be for level 5 autonomous cars: Which country the car is in so it identifies the fact that swerving one way in a particular country will take it into oncoming traffic rather than say the road shoulder. Also, the type of ground it is on so it can predict how much grip it will have e.g. tarmac, gravel, paved etc. The condition of the car’s tires and brakes. Whether the car’s occupants have fastened seatbelts. Whether the car is transporting dangerous cargo that could spill or explode, proximity to hospital or emergency rescue, damage to property such as houses and buildings, and more. All of this data influences the probability of an accident as well as expected harm, both of which are needed in selecting the best course of action.
4. Testing
To gain a license as a company to have a self-driving car on public roads with pedestrians and other vehicles etc., the vehicle must have driven millions of miles without an accident. The biggest companies have achieved this and have begun testing their level 2-4 autonomous cars on public roads. An example of one of the biggest companies is Waymo, the former Google self-driving project, who have driven eight million miles on public roads. Four million of these miles have come in the past eight months. Waymo’s fleet of self-driving vehicles are now logging 25,000 miles every day on public roads with 600 self-driving Chrysler Pacifica Hybrid minivans. Using AI simulation programs on a computer, Waymo has “driven” more than 5 billion miles in its simulation, according to the company. That’s the equivalent to 25,000 virtual cars driving all day, every day. Waymo is getting closer to launching a commercial driverless transportation service later this year. About 400 residents in Phoenix Arizona have been trialling Waymo’s technology by using an app to hail self-driving Chrysler Pacifica Hybrid minivans. The company says it plans to launch its service later this year.
Toyota have also built a new 60-acre test center which is set up like a city in a remote part of America to simulate realistic Road situations that are too dangerous to perform on public roads. This will allow the company to really push the boundaries in autonomous car testing and rapidly advance capabilities. The site will include congested urban environments, slick surfaces, and a four-lane divided highway with high-speed entrance and exit ramps.
The biggest Impacts on the industry from all this testing are the huge costs in investment to build infrastructure to support the research as well as the costs to increase the rate of technological advancements
The cars drive around 24 hours a day to rack up the miles
They have to demonstrate to the authorities that the city situations are realistic and not memorised/
5. Business and commercial use of self-driving cars
This rapid progression in the world of self-driving cars has encouraged many partnerships and agreements in the industry to develop the technology as fast as possible. The Japanese carmaker Toyota is set to invest £388m in disruptive company Uber as the two companies expand their partnership on the development of self-driving cars. It strengthens an existing relationship in a bid to catch up with rivals in the race to design and produce autonomous vehicles for the mass market.
As I previously mentioned, Uber halted its autonomous cars from testing on the roads and closed its self-driving testing hub in Arizona. Under the latest deal with Toyota, Uber will combine its autonomous driving system with Toyota’s Guardian technology, which offers some automated safety features such as lane-keeping but does not enable a vehicle to entirely self-driven. The end goal is to implement this technology into Toyota’s Sienna minivans, to be used on Uber’s ride-hailing network from 2021.
Everything about the commercial and business use of these cars his about cost per mile, removing the cost of a driver in all these vehicles will drastically reduce this cost per mile. -aobut 20p per mile rn, should go down