Driving into the Future of Autonomous Cars
Math 89S: Mathematics of the Universe
Professor: Dr. Hubert Bray
November 1st, 2016
Every year, 1.3 million people die in car crashes around the world and 37,000 of those deaths occur in the United States (3rd most after India and China). In the United States 8,000 of the deaths that occur involve drivers between the ages of 16 and 20. Furthermore, in the United States, 2.35 million people are injured or disabled per year by car crashes. There have been multiple attempts to decrease the number of accidents that happen on the road. These solutions range from apps that prevent drivers from using their phone to companies creating cars with “automatic breaking”. These solutions, however, have failed to significantly decrease the number of car crashes that occur as they do not address the main problem in all car crashes, the driver. Ninety percent of all motorized vehicle accidents are caused by human errors which is why many experts believe that the most effective approach to decreasing the number of accidents that occur on the road would be to create autonomous cars. In this paper, I will outline a timeline of autonomous cars, describe their functionality, discuss their economic feasibility, and evaluate the benefits and disadvantages to using autonomous cars in the future. 
What are Autonomous Cars?
One of the key concepts to understand when dealing with autonomous cars is what it means to be “truly autonomous”. A truly autonomous vehicle is a car that is capable of operating without a human. Currently, we have vehicles that are close to being truly autonomous (Tesla Motors is currently the frontrunner), however, we are still years away from creating cars that are able to drive on their own.
Autonomous Cars: An idea from the Past?
Before understanding how current day autonomous vehicles work, we need to understand how they have been developed up to this point in time. One common misconception that people have about autonomous cars is about when they first came to exist. Many people believe that autonomous cars have only been relevant for the past decade however, car manufacturers have been developing autonomous cars since the 1920s. Below is a brief outline that examines the history of autonomous cars through the past century:
1920: In the 1920s, a 1926 Chandler was equipped with an antenna to create the Linrrican Wonder. The technology behind the Chandler was quite simple, a second car would follow the Chandler and send radio impulses through antennas to the Chandler’s circuit-breakers. These circuit-breakers were linked to small electronic motors that would ultimately direct the movement of the Chandler(Mario).
1950: In the 1950s, RCA Labs was able to create a miniature car that was controlled by wires that were laid on the floor of a laboratory. This ultimately evolved into a full-size system in Nebraska. Inside the pavement, detector circuits had been buried inside that ultimately send impulses to navigate the car and determined the speed of other vehicles on the road(Vanderbilt).
1980-90: It was in the 1980s when we first began to see glimpses of autonomous cars that were able to function without helper devices. In the start of the decade, we saw the emergence of the vision-guided Mercedes Benz which was able to achieve a speed of 39 miles per hour on an empty road. A few years later, the ALV project (Autonomous Land Vehicle) used LIDAR and computer vision to control a vehicle at speeds of 19 miles per hour(Vanderbilt). By the end of the decade, Carnegie Mellon University was able to use neural networks to steer and control autonomous cars and is now the foundation of contemporary control strategies(Mario).
2000-Present: In the start of the 21st century, we begin to see the formation of government-funded programs geared towards creating autonomous vehicles. It is not until 2010 that we begin to see major automotive and tech companies venturing into the field of autonomous cars. In 2010, VisLab (Italian based) was able to test run an autonomous car (9,900 mile run) which marked the first intercontinental land journey by an autonomous vehicle (Gerla). Furthermore, in 2010, the Institute of Control Engineering of Technishe Universitat Braunchweig developed the first autonomous car that was able to drive on the streets of Germany (with traffic). After 2010, multiple companies such as Google, VisLab, and Nissan were able to create cars that were able to drive autonomously but were still not ready to be driven on the road as they lacked the ability to follow signs and other road/driving protocols(Vanderbilt). Finally, in 2015-16, Tesla Motors released autopilot technology that ultimately allowed their cars to drive autonomously. Throughout the past year, Tesla released multiple software patches that increased the autonomous abilities of their cars (ex: allowed Tesla vehicles to self-park). This month, October 2016, all of Tesla vehicles are now at an SAE level of 5. Tesla still, however, has not yet been able to release full self-driving vehicles as they need to continue testing their cars. Tesla recently stated that they plan to reach full self-driving by the end of 2017 to the start of 2018.
The Functionality of Autonomous Cars:
Before delving into how autonomous cars function, it is important that one understands how autonomous cars are classified. The SAE International (Society of Automotive Engineers) developed a classification system that is based on the “amount of driver intervention and attentiveness needed rather than capabilities of the vehicle/autonomous system” (Stephen). Below is a table published by the SAE that describes the different levels:
This table may be difficult to understand, especially for those with no driving experience, so I have created a table on the next page that summarizes the different levels without the mechanical jargon:Level 0 / An automated system does not control the vehicle. The system may issue warnings or alerts. Ex: An alarm that beeps when a car is near an object when being reversed
Level 1 / An automated system where the driver should be ready to take control of the vehicle at any time. Ex: Cruise Control
Level 2 / An automated system where the system is in charge of driving, steering, and braking and the human is required to detect objects. Ex: low level automated vehicles
Level 3 / The driver does not have to do anything when driving in a known environment such as a highway. The driver should still be ready to intervene if need be.
Level 4 / The driver does not have to anything in almost all environments except for the case of an outlier (bad weather)
Level 5 / An automated system where the only human interaction required is to set the destination point.
How do Autonomous Cars work?
Now that we have a general idea of how autonomous cars are developed, we should now be able to better understand how autonomous cars work. For this paper, the main vehicles that we will be looking at will be cars from Google and Tesla as these companies have been the most successful in the autonomous driving industry. Below is a detailed description of how Google and Tesla autonomous driving systems work:
One of the main differences between Google and Tesla’s automotive systems is Google’s usage of Light Detection and Ranging technology (LIDAR). Lidar is a surveying method that measures the distance to a target by illuminating that target with a laser light(Nicholls). In Google’s car, the LIDAR technology consists of 64 lasers that spin at 9000 rpm, ultimately producing a 360-degree view. The survey of the environment created by the LIDAR is then converted to create a 3D graphical representation of the surrounding environment so that obstacles in the environment can be viewed in real time. Once the 3D representation is created, the car uses its attached sensors to determine the position of the different obstacles and objects present in the environment so that the car can move accordingly. The car also has a video camera (attached to the console) that enables the car to see pedestrians and stoplights that are difficult to catch with the LIDAR. One of the advantages of the Google Car is its ability to sync Google Maps, one of the premier mapping services available today. Google uses Google Maps to help determine the path that the car will follow and also the cars current position(Times). Since correct positioning is crucial for an autonomous car, there is also a position estimator on the left wheel of the car that detects the speed of the car which ultimately results in the system having a more accurate position of the car. The graphic below illustrates the process described above in a more detailed manner:
In addition to the technologies described above, Google is also attempting to implement a new technique known as vehicular communication. Vehicular communication is a method that will allow cars to communicate with each other on the road through cloud servers. Vehicles would then be able to pull this information down from the server, and thus have a better understand of how traffic flow is in certain areas which would ultimately aid them by helping them determine the most efficient path (quickest and safest) to go from one place to another.
As mentioned above, Tesla and Google have different mechanisms to create a visual representation of a car’s environment. Both Google and Tesla revolve heavily around sensors, however, the main difference as mentioned above is Google’s usage of LIDAR. Tesla’s autonomous system (also known as Tesla Autopilot System) is composed of multiple sensors that are placed around the car to help it understand its environment. The image below shows how a driver sees the Autopilot system:
In terms of hardware, Tesla’s vehicles currently “includes a forward radar, a forward-looking camera, a high-precision digitally-controlled electric assist braking system, and 12 long-range ultrasonic sensors placed around the car” (Thompson). These ultrasonic sensors are strategically placed around the vehicle so that the car is able to sense when something is too close and gauge the appropriate distance so that it can slow down. One important detail to understand about these sensors is that they can report inaccurate information if there is something that is interfering with them. We saw this in the fatal Tesla accident, where a man died in a car accident because of an “issue with the autopilot system”. While the exact reasoning is unknown, many experts believe that the Tesla sensors were affected by the brightness of the sun which causes the sensors/radar to believe that the truck was actually an overhead road sign because of its large height. While this is a problem, we also need to be reminded that truly autonomous cars do not currently exist (no car is at an SAE level 5 right now), so a driver using Tesla’s Autopilot needs to be ready to take over at all times. Going back to the sensors, the image below shows the “circle” that is created by the sensors placed around the car:
Tesla’s autopilot system ultimately takes the data inputted by the sensors and cameras and creates a digital representation of its surroundings (stationary and moving objects). The image below shows an example of how the digital representation looks for a car:
As you can see from the image, the digital representation created is quite similar to the image created by Google’s car which ultimately emphasizes how despite using different methods to learn, both autonomous systems yield similar graphics to display the information that they have learned.
In addition to learning about its surroundings, one of the other important parts to an autonomous system is its decision making algorithm/ability. Currently, no car is truly autonomous as they still have drivers, however in the near future, these cars will be required to make decisions on whether they should stop, slow down, or swerve when in an emergency situation. While Tesla and Google have not yet fully developed autonomous cars, many people are already discussing the ethical dilemmas that will result from a computerized system making a decision about life or death. While there will be some ethical challenges, I believe that the challenges are being over-exaggerated due to the lack of knowledge people have about these systems. One common misconception about the decision making process is that many people believe the system will be a series of if/else statements (ex: If a person is in front of Car- turn right). In reality, however, these machines will rely on machine learning and pattern recognitions to make decisions ultimately allowing them to mimic the decision making process of a human(Galceran). By using the elements of AI, autonomous cars will be able to learn and analyze how “good” drivers drive and ultimately replicate their driving ability. Once autonomous cars reach an SAE level of 5, the next issue is the “Trolley Problem”. Currently, engineers from both companies are saying that they are teaching their systems to hit a stationary object whenever there is a chance that it might hit a human being. While this is easy to program, the real problems begin to occur when the system has to decide between hitting one person vs another. Engineers have stated that through machine learning, cars will be able to detect which path would lead to the least amount of distress and follow that path(Galceran). Overall while these problems must be considered, one needs to realize two things, the first is that these hypothetical situations rarely occur in the real world and the second is that a computer system will most likely make better decisions (in terms of what to do in an accident/situation) in comparison to a bad driver.
Currently, the safest small car to buy is a 2018 Honda Civic which has an MSRP of $18,600. While Google has not yet released its car in the market, the current cost of just LIDAR is $75,000 per vehicle which is already almost four times the price of a Honda Civic(Stephen). When adding all other costs, the current price of a car from Google is looking to be over $100,000 dollars making it virtually unaffordable to almost all Americans (average car budget for a person living in the U.S is $17-$33K). It is because of this absurd price that Google has stated numerous times that the price of LIDAR is going to drop significantly over the next decade ultimately making the car more affordable. Tesla on the other hand is now producing cars with autonomous systems that have a starting MSRP of $30,000 as they already have cars with elements of autonomous driving (SAE level 2-4) that go for anywhere from $66,000 to $110,000. As one can see, it is far more affordable to buy a Honda Civic than it is to buy an autonomous car, however as advancements in technology continue, the cost of manufacturing these cars will decrease.
Overall, while we still are far away from having truly autonomous cars, we have made great strides in creating cars that have some elements of autonomous driving. Companies such as Google and Tesla are the frontrunners in this field, and other companies are trying to join the movement. Just recently, rumors about Apple purchasing McLaren began to circulate ultimately showing their interest in also joining the autonomous driving industry. It is because there are so many different companies involved in this industry, there are a variety of different approaches to creating an autonomous car. Some of the common elements that can be found in all of the cars developed up to this point are computer vision, the usage of a variety of sensors, cameras, and forms of neural networks. Decision-making is still an unclear process as fully autonomous cars (SAE Level 5) have not yet been released to the public, however, the usage of artificial intelligent learning methods looks to be an efficient way for these systems to make decisions. Overall, I believe that autonomous cars, over the next few decades, will become cheaper, and safer than the cars we currently have today and will ultimately help us drive into a safer future.
Galceran, Enric, Alexander Cunningham, Ryan Eustice, and Edwin Olson. "Multipolicy Decision-Making for Autonomous Driving via Changepoint-based Behavior Prediction." Robotics: Science and Systems XI (2015): n. pag. Web. 1 Nov. 2016.
Gerla, Mario. "Internet of Vehicles: From Intelligent Grid to Autonomous Cars and Vehicular Clouds." (2014): n. pag. Web. 1 Nov. 2016.