Transportation has been a necessity since the start of human civilization, where narrow dirt paths evolved into wide 10-lane highways. Yet for thousands of years, the requirement for a human operator has stayed. We’ve built faster cars, safer roads, more powerful engines, bigger planes…, but the “human factor” has remained unchanged. With the recent, called by many, “AI Revolution”, our inadvertent mistakes are reduced using computer algorithms and sensors, and it aims to take over human control in transportation soon.
Historical background
The Industrial Revolution started in the UK in the XIX century, and until then, most machines were powered by human power or horsepower. With the rise on popularity of steam engines, horses have yielded and steam power ended up dominating the industry, but humans were still involved to control these machines, which are inanimate.
The first shift towards automation was with James Watt’s adaptation of a “centrifugal governor” to his steam engines, which allowed them to control their own speed by regulating fuel flow.
In air transport, planes required constant monitoring and focus of the crew to maintain heading and thus created fatigue. To remedy this, in 1912-1914, the first autopilot was created by Lawrence Sperry. It connected a gyroscope to the plane’s control surfaces on the wings hydraulically. Automotive engineer Ralph Teetor also built the first modern cruise control system for automobiles in the 1950s. But these systems aren’t smart: They can maintain a speed or heading but not avoid obstacles.
In the late 1990s, two semi-autonomous car prototypes were developed by the Germans as part of a larger European project: “VAmP” and “VITA-2”. These have driven over 1,000Km in highways and heavy traffic successfully, and they performed tasks like changing lanes, and reacting to spontaneous situations.
In the modern days, with the creation of sophisticated AI learning algorithms, automotive giants like Tesla, Honda and Mercedes-Benz started commercializing partially autonomous vehicles, with Tesla leading the market.
How does AI work (in transport)?
AI is a branch of computer science. It specializes in creating dedicated systems that perform tasks that typically require human intellect. Unlike traditional robots or computer software, AI “thinks and learns” from experience overtime.
Machine Learning is the principle of most modern AI, where instead of a developer writing manually each scenario in code, the system “learns” and adapts with all the data it’s fed. This data can consist in anything, like reports or books, driving footages, traffic patterns…
It usually uses a process called Deep Learning to train Neural Networks. These networks are layers of thousands of mathematical functions, where it senses it’s environment, thinks, then takes a decision, inspired by the human brain.
AI on the road
To know what actions the AI needs to take, multiple sensors are used, usually in a redundant system, so if one fails, the rest still work fine.
High-resolution cameras are placed around the entire vehicle, which detect things like traffic lights, but they can be blinded by rain, fog, or the sun.
LiDAR is also widely used, it’s a device installed on the roof of the vehicle, and it spins, firing millions of laser pulses, which detect the distance to a point, creating a 3D view. LiDAR can know milimetrically where an obstacle is, but it can’t detect traffic signs.
Other sensors like ultrasound and gyroscopes have been used in some models and prototypes.
It’s worth mentioning that in the industry there is a major disagreement between companies, where Waymo and Uber believe LiDAR is necessary, Tesla uses a camera-only vision. Elon Musk says that since humans only use vision to drive, AI should also be.
Today, one of the related projects that is believed to have the most future, are “Robotaxis”. These, ran by companies like Waymo and Uber, offer taxi services with driverless cars across the world, including European cities like Madrid. Beyond taxis, brands like BMW and Tesla are still trying to push the industry into private vehicles.
Some US companies have also developed trucks with AI, that communicate with each other creating a “convoy”, that reduces fuel consumption.
Last year, driverless electric buses were also tested in the dense roads of Barcelona and Madrid, designed to work “on demand”, meaning that they will automatically adapt the route to the passenger’s needs.
Rails and Automation
Railways have been leading the “AI revolution” for a long time. Rail lines are divided into small sections called “Blocks” of about 500m to 1Km each, where only one train can be in a section at a time. Modern railway systems aim to eliminate this.
The first example of automation in a train is the ATO system, developed around the 1960s, “Automatic Train Operation”. Like its name suggests, it drives the train automatically, but it still requires for a driver to power up the train, open/close doors… ATO still follows the traditional “Blocks” and is used by most metro and suburban systems nowadays, along ETCS. One of the first railways to use this system is the Barcelona Metro in 1961.
ETCS (European Train Control System) is a security system developed for the European context in rail transport, where the train’s speed is constantly monitored and emergency stop is applied to the train when needed. Newer ETCS models are packed with a system that removes the traditional “Blocks” and allows for trains to drive in shorter distances one to another, making the rail lines have a larger capacity and safety. These newer models are more used in intercity and high-speed rail rather than urban networks. ETCS is also designed to work alongside with other European railway systems, hence why it’s massively used.
Full automation is now achieved by CBTC (Communications-Based Train Control) in the Barcelona and Madrid Metro, where like ETCS, eliminates the necessity for “Blocks”, but also adds full AI control on the trains. These trains don’t have a driver’s cabin, and are driven exclusively by AI, with only a small dashboard for emergency operation. It relies on communicating with other trains and the dispatchers, and the system knows where the train is in real-time.
The ethical side
Why can’t AI take over aviation yet?
Autopilot is just a tool created to reduce fatigue in crew and offer a more comfortable and pleasant ride to travellers. Pilots can select the altitude, heading, speed and pitch of the aircraft, and it will make the plane fly level, but it won’t make decisions by itself.
AI does millions of calculations per millisecond, and it costs large amounts of power for all that computing to work, and it will cost more fuel, because electricity comes from the plane’s engines.
In planes, unlike in cars, up to hundreds of extra parameters must be considered because of the added complexity of “3D” space.
The dense airspaces, exceptionally in Europe, China and the US, require a steady communication between pilot and controllers. While this has been achieved by the company Cirrus, it’s available only for emergency landings only, in case there is something wrong with the pilot.
The Trolley Problem
Would a self-driving car whose brakes failed swerve and crash into a wall to save 5 children that were crossing the road, potentially killing its own passenger?
This is the classic “The trolley” problem but applied to AI and vehicles. It’s been a subject of debate in car manufacturers. Self-driving vehicles are designed to do their job: Obeying traffic laws and protect the passenger. It won’t do split-second decisions on where to swerve to; it will always first try to hit the brakes and avoid obstacles at all costs.
Will we need driver’s licenses in 20 years?
“The question now is not if we adopt new technology, but how boldly we choose to do so. Success is not about how many vehicles we deploy, but how intelligently they are integrated into the system.”
Robert Falck, CEO of Einride
It’s not surprising that this is a really asked question about the future of transport. By 2046, it’s highly expected to have started a transition towards “unattended” vehicles (vehicles with no pedals nor steering wheels). This means that, for the next generation, learning to drive may become learning to operate: People will license on how to handle complex questions to AI instead of driving a car.
Governments like the EU can impose safety regulations to manufacturers, potentially imposing deep software inspections to ensure vehicles don’t have a defective AI, which can cause lots of accidents.
In 20 years, “Do you have a license?” will ultimately be “Is your vehicle licensed to carry you?”
