How we can save billions of dollar every year, and why we still haven’t done it yet.

Original article was published by Carina D’Souza on Artificial Intelligence on Medium

How we can save billions of dollar every year, and why we still haven’t done it yet.

…and how to reduce cancellations, delays, collisions, carbon emissions, and flight time

Picture this: You’re on a road trip and can take one of two paths to get to your destination.

The first path is a road with a lot of detours, and delays along the way. The second path is a straight road to your destination, which saves you more time, and requires less gasoline to get to the destination, compared to the first road. In addition, the second road is much more easy to navigate, since it’s a straight path, meanwhile the first road is crowded and full of detours.

The second road is simply much more efficient, saves time, and requires less money and thought to operate.

Knowing this, who wouldn’t choose the second path?!

…well, apparently, all of America. 🤯

On average, around 7,000 aircraft fly over the United States per day. For the past 40 years, we have relied on the same air traffic control (ATC) system to direct flights, even though modern technology can permit us to save more money, more time, and speed up flight time.

The problem

While the opportunity to save money, time, and the environment is a good incentive to implement our solution, there is also a fundamental flaw in our current ATC system.

Currently, ATC controllers are responsible for planning our every single flight path, every day. This repetitive process is prone to human error, and many planes crash every year due to poor communication between ATCs and pilots, or just poor flight planning. A study conducted by NASA found that almost one-fifth of ATC controllers made significant errors due to chronic fatigue due to lack of sleep from their busy sleep schedule. ATC controllers are also more prone to stress-related illness, such as heart failure.

A peek inside the current ATC system

This system also cannot cope with the increasing demand of plane flights — heavier traffic and constraint control capacity is leading to big increases in flight delays and cancellations. In the US alone the number of flights delayed or cancelled by air traffic control rose 69% in the past year.

Delays and cancellations cost the US economy over $28 billion in 2018 alone. Delays and cancellations can be eliminated entirely with proper flight planning.

We’ve all been here before.

This is insane, not only are we operating under an outdated and inefficient system, this system can cost peoples’ lives, and lead to huge, several billion dollar, losses.

Our solution

To solve this problem, we use a recurrent neural network (RNN) and deep reinforcement learning.

What is an RNN and Deep Reinforcement Learning?

An RNN is a type of neural network which uses data points in a sequence to make better predictions. They do this by taking in input and reusing the activations of previous or later nodes in the sequence to influence the output. Our RNN will take real-time data (such as weather conditions, flight position, fuel remaining, etc.) and keep refreshing that.

How an RNN works; Taking in data points in a sequence to predict outputs

Deep Reinforcement Learning is a combination of reinforcement learning, learning by adjusting actions based on feedback to maximize a reward, and deep learning, training from a dataset and then applying the learnings to a new data set. Our deep reinforcement learning algorithm will train off of geographic specific data to optimize aspects such as saving fuel, time, avoiding collisions, etc.

How Deep Reinforcement Learning works

These two networks will work in conjunction to optimize for the most cost effective, the quickest, and the safest flight route.

The algorithms will co-function will a human ATC controller, who will have the final verdict on the plane’s path, and will relay the information to the pilot of the plane. This is to make sure that the algorithm doesn’t malfunction, or recommend a faulty path.

Based on our research, we discovered that this could result in over a 12% saving in fuel consumption per flight.

On average, a 10 hour international flight on a Boeing 747 requires 36,000 gallons of fuel. Fuel costs approximately $7,070 USD for 3,500 gallons of jet fuel. If we can save the plane from using 12% of it’s fuel, that means we could save 7,200 gallons of fuel, which is approximately $15,000 USD per flight.

Not to mention how much time will be saved per flight, and the number of new flights that will be able to be created because of the saved time, and the billions of dollars saved from avoiding cancellations.

Reducing the time in the air per flight also has huge implications for the environment as well — aircrafts in the USA account for 3% of the country’s greenhouse gas production.

Beyond Co2: The Impact of Air Travel on Climate Change.