Transportation Planning Casebook/The emergence of Self-Driving Vehicles

Suggested Readings

 * USDOT Connected Vehicle Research
 * Considerations for the Design of Automated Urban Transportation Systems
 * PRT Consulting Video on Masdar City, UAE
 * Transport Modeling by Multi-Agent Systems: A Swarm Intelligence Approach
 * Guided by Computers and Sensors, a Smooth Ride at 60 Miles Per Hour

Early Motor-Driven Vehicles
The modern automobile has roots going back to the 17th century and models built by Ferdinand Verbiest for the Chinese Emperor. From that toy, it took a century or more for full sized steam-powered vehicles to emerge. Around 1800 further advances came as steam power was replaced by internal combustion engines. Still, automobiles did not emerge until the late 19th century despite the vast rise in steam- and coal-powered engines used for railroads.

1885 brought the first modern automobile, that is a vehicle specifically designed to be powered by a motor, from German engineer Karl Benz. Interestingly, the Benz Patent-Motorwagen was a three-wheeled vehicle (one leading, two trailing) unlike current four-wheeled vehicles. From 1885 to 1920, the automobile saw a number of advances, including the transition to gasoline power (by around 1910).

Precursors for Autonomous Vehicles
Prior to 1920 the majority of automobiles were started by a hand-crank mechanism. Due to the erratic behavior of motors during startup, manual starters often injured early car owners. As early as 1896 electric starter motors were being developed and deployed on automobiles a by 1920 were standard on new vehicles. This is the first of many precursor technologies required for the autonomous vehicle to become a reality.

Development of additional precursors emerged just before and during World War II with the automatic transmission (1930s) and shortly after the war with cruise control (1940s). With these, a vehicle could drive at a predefined speed along any number of grades or curvatures with drivers only necessary for steering. However, the cruise control could not be controlled except by a human user until the 1960s-70s. As electronics became cheaper, smaller, and more sophisticated the mechanical cruise control was replaced by a computerized version capable of interacting with other vehicle control systems.

Current Technologies and Advances
With the rise of the computers and sensor technologies (sonar, radar, etc.), modern vehicles have been equipped with a number of automated features to assist drivers. Among the most important in terms of transitioning toward autonomous vehicles are automatic braking, lane departure detection, blind spot warnings, GPS navigation, and automatic parallel parking. These types of systems automatically detect the presence of other nearby vehicles and/or lane lines to avoid collisions and complete tasks difficult for a human driver to perform quickly and reliably.

Further reading on these technologies can be found here.

Autonomous Vehicles in Science Fiction
As cars grew into the mainstream, authors and filmmakers stretched the boundaries of vehicle technology and offered some ideas as to the future of personal vehicles. As early as the 1950s, authors like Philip K. Dick and Robert Heinlein included automated cab systems in novels and short stories. These fictional systems generally resemble the technologies being developed today: fully autonomous vehicles, voice control systems, automatic navigation

More modern depictions appear in film and television programs such as Knight Rider, Minority Report, and I Robot.

Google’s Self-Driving Cars
The Google’s self-driving car project involves using artificial-intelligence technology to cars drive automatically without interventions of human drivers. The project is led by Sebastian Thrun, who is the director of the Stanford Artificial Intelligence Laboratory, a Google engineer and the co-inventor of the Street View mapping service.

Technologies
The automated Google cars are adapted from conventional cars. Originally Google had a fleet of 7 self-driving cars, including six Toyota Prius hybrids and one Audi TT. In 2012 Google adds Hybrid Lexus RX450h to the fleet.

The idea is to use artificial-intelligence software to sense the surrounding environment and simulate decisions made by a human driver. The process is realized through video cameras, radar sensors, a laser range finder equipped on the cars, and detailed maps information provided by Google’s data centers to help navigate the cars.

The cars make announcements during automated driving like “approaching a crosswalk” or “turn ahead” to alert the driver if a master control system detected anything amiss with the various sensors.

The sensor mounted in the center of the car is called Lidar, or Light Detection and Ranging. It provides a continuously updated three-dimensional map of the world at centimeter accuracy extending for more than 230 feet around the car.

The car has four standard automotive radars with less resolution and greater range, three in front and one in the rear. Inside the car there is a high-resolution video camera positioned next to the rear-view mirror, which is used to detect street lights and moving obstacles like pedestrians and bicyclists. The car also has a GPS receiver and an inertial motion sensor.

There are three ways for the human driver to to gain control of the car: hit a red button near his/her right hand, touch the brake or turn the steering wheel.

Testing Google Cars
The automated cars are adapted from ordinary cars. Google equips the cars with technology that enables them to drive automatically. Initially Google ran the self-driving car project in stealth mode, and later it revealed it to the public in late 2010. By then, all the self-driving cars have driven 1,000 miles without human intervention and more than 140,000 miles with only occasional human control. The cars drove through places such as Golden Gate bridge, Pacific Coast Highway, Lake Tahoe, Google’s Santa Monica office, and even Lombard Street, one of the steepest streets in the United States which consists of many tight turns.

The tests continue since then. By March 2012 the tested mileage was 200,000 miles. Google announced on August 7, 2012 that the self-driving cars completed more than 300,000 miles of testing, without a single accident under computer control.

So far, the automated driving tests follow several procedures. At the beginning of the tests, Google briefs local police on its work, then sends out a driver in a conventional car to map the route and road conditions, recording features like lane markers and traffic signs. The mapped information will be annotated by a group of software engineers to make sure the road signs, crosswalks, street lights and unusual features are all embedded. All the information is then processed and stored by Google’s data centers. After this, an automated test drive will be conducted along the same route. While the car is driving automatically, it records changes as they occur and updates the map. The whole process, including the manual driving and then the automated driving, is known as SLAM, or simultaneous localization and mapping, which creates and updates a map while the car is located within the map.

The self-driving cars are not unmanned. In the cars will be a trained safety driver behind the wheel who can take over the car when necessary, and a trained software operator in the passenger seat to monitor the software.

The Laws Regarding Self-Driving Cars
The advent of automated cars poses thorny legal issues. The current law requires a driver behind the wheel at all times. Also many motor vehicle laws presume to have a human being operating the cars. As Bernard Lu, the senior staff counsel for the California Department of Motor Vehicles said, “the technology is ahead of the law in many areas.”

Originally Google tested their self-driving cars on public roads in California. Although it was not officially allowed by law, it was determined legal by California’s motor vehicle regulations because a human driver was behind the wheel and could override any error.

Nevada is the first State to authorize self-driving cars for the state’s roads. The law went into effect on March 1 2012. Florida became the second state to pass a bill allowing tests of self-driving cars in April, 2012. California is the third state to enact autonomous vehicle legislation (September 2012).

Accidents
Google’s first self-driving car crash happened in August 2011. One of Google’s Prius rear-ended another Prius. But Google says the car was not in auto pilot mode at the time of the collision.

Costs
Besides the costs for a conventional car, a Google’s self-driving car needs to carry about $150,000 equipment. The Lidar on top of the car alone will cost $70,000. Apparently it is too expensive for consumers today.

Potential Effects from Google Cars
Improving safety is the major goals for the project. Google thinks that robot drivers have advantages over human drivers because they have 360-degree perception, react faster, and do not get tired or distracted. So the use of self-driving cars will potentially reduce the road traffic accidents and save lives. Google cars can drive closer to each other on the road than manually driven cars, (“highway trains of tomorrow”, as in Google’s words. Thus they can improve road capacity (double the capacity of roads by Google’s estimation, increase time efficiency, and reduce congestion.

If the Google cars reduces car accidents, then the cars can be built lighter. Reduced fuel consumption is likely to result from reduced congestion as well as lighter cars.

Google cars will also probably change the way people use cars and promote car sharing. When the cars eventually do not need anyone behind the wheel, they can be summoned electronically, so that people could share them. People can simply make calls anywhere and wait for the automated cars to pick them up. The car sharing will reduce the need for cars and parking spaces.

Problems That Needs to be Solved
Currently, the design team of Google cars hasn’t been able to train the cars to recognize the hand signals made by a human traffic cop or crossing guard. The team also need to work on the interpretation of temporary construction signals and other tricky situations that many drivers encounter.

To make the self-driving cars truly safe, the computer systems in the car need to be very reliable. Today’s personal computer is not reliable enough because they crash occasionally and is vulnerable to virus infection.

Liability is also a potential issue with driverless vehicles. In the event of an accident, who would be liable — the person behind the wheel or the maker of the software?

Future
There are numerous ideas that are being explored for the future of vehicles. Some systems would be fully automated, while others would simply create smarter cars. Here are some examples of ideas that are currently being researched, or at least thought of.

Vehicle-to-Vehicle (V2V) Communications
V2V is a system where cars within a certain vicinity of one another communicate via wireless transmission. A vehicle sends a Here I Am message to the vehicles within the range of its transmission. The Here I Am message is a simple data message that sends out the vehicles location, speed, and trajectory. This data is then received and processed by the computers in other vehicles in the area. The computers then make decisions based on the data it receives from the set of cars in the area to minimize risk. The ultimate goal would be for all vehicles on the road to have V2V capabilities, so a wealth of data was available for all the vehicles. All of the technology needed to implement the system is already available. A GPS system can provide the necessary data, many newer cars have wi-fi capabilities that can transmit a data signal, and as discussed earlier some cars have driver assist capabilities that can override a drivers action to prevent a collision. Cars need not have driver assist capabilities though for V2V to work. Warning systems and alerts can also serve to aid the driver. The USDOT is currently undergoing research on the possibility of a V2V system, and the overall safety benefits it would produce. A V2V system has the benefit of allowing drivers to still have the same freedom to drive that they currently enjoy, while limiting dangerous collisions. A potential downside is people's apprehensions of sharing data of their driving to others.

Vehicle-to-Infrastructure (V2I) Communications
V2I is a similar idea to V2V in that cars are still transmitting travel data, but instead they are communicating with infrastructure such as traffic signals and digital road signs. The ultimate goal of V2I is to better inform drivers while minimizing the infrastructure needed. Also like V2V, the technology is available and only need be added to vehicles and infrastructure. V2I is seen as a complimentary system to V2V and is also currently being researched by the USDOT. As of now, the USDOT sees this technology being used as driver assist, but it is possible that it later evolve into driverless.

Automatic Guided Vehicles (AGV) and Personal Rapid Transit (PRT)
AGV is an automated system where the vehicles are operated on a set path, typically on rails or by the use of magnets. AGVs are becoming very common in the manufacturing industry where they have replaced human operated modes of transporting materials such as pallet jacks and forklifts. The idea of using AGV for human transportation has come about in science fiction literature and films, and there have been recent attempts to implement such a system in a select few places. In Masdar City, UAE just outside of Abu Dahbi, there was an attempt to put in a Personal Rapid Transit (PRT) system instead of the traditional roadway system in this new build city. The PRT system had a series of driverless electric cars that were guided by magnets under the roadway. The cars are not restricted in their travel path and can take riders anywhere in the city. A rider would simply arrive at a station, and use the platforms computer to call for a car. The car would arrive and the rider would input the destination. Riders also have the added benefit of public transportation that they don't have to share with the public. Ultimately, the PRT system of Masdar City was scrapped in a cost cutting move, with the exception of one stretch from a light rail station to the University.

Swarm Intelligence Transportation Systems (SITS)
SITS is a relatively new concept. It is a system based off the behavior of social insect species such as ants, bees, and termites. Sometime also called Collective Intelligence, it refers to the idea that each individual agent acts and moves in the optimal manner for the whole. The current research being done in SITS is in an effort to improve the current transportation systems through better modeling and analytical processes. The future of swarm intelligence is vast. There is the idea of a centralized controlling network of super computers which would control all transportation movements, and optimize the efficiency. This concept is still a ways off, but will likely be brought up sometime in the future. A drawback to this system is with everything completely automated, eliminating all drivers freedom to drive, but the plus side is that everything is automated, so virtually all accidents would be eliminated