Year 2036: Aboard an electric aircraft in an autonomous ecosystem

Out by the gate, a small electric regional aircraft is waiting.

I haven't booked the trip myself. When the meeting in Helsinki appeared in my calendar, my AI agent handled the rest. Flight, hotel, and rental car were booked simultaneously, based on my preferences and with the right margins before and after. I'd first thought about taking the ferry, but the agent's calculation showed a lower climate footprint with the electric flight this time.

Since booking, the departure has changed once. I noticed it only when my calendar updated. The new departure still left enough margin before the conference, while also letting the airline's system make better use of capacity.

As little time on the ground as possible

Check-in goes quickly. When ground time needs to be kept as short as possible, every step can become a bottleneck. Identification and boarding happen almost entirely automatically, while anomalies are flagged for human review.

While we travelers are shuttled through, the aircraft are being charged according to a complex schedule that weighs several factors against each other. Multiple planes need power at the same time, but have different departure times, different amounts of charge remaining, and different margins.

A plane departing in 35 minutes might need to be prioritized over one that has two hours to spare. At the same time, the system can anticipate that several aircraft will soon land and start competing for the same limited power. In that case, it may make sense to charge more now, before the load increases.

But charging as fast as possible isn't always the best choice. Repeated fast charging wears on the batteries, and data such as temperature, voltage, and past charge cycles together paint a picture of the battery's health over time.

So the goal isn't always to charge to full as quickly as possible, but to give each aircraft the right amount of energy at the right time — without unnecessarily shortening battery life or creating a new problem elsewhere in the system.

Autonomy continues in the air

Once we're airborne, the route is of course planned. But it isn't set in stone.

Winds change. Air traffic changes. The estimated arrival time changes. Altitude, speed, and flight path may therefore need adjusting mid-flight to use energy as efficiently as possible.

The more autonomous aircraft sharing the airspace, the more complex the coordination becomes. From a distance, the movements could almost resemble a flock of migratory birds: each individual follows its own path, while reacting to the movements around it and adjusting direction and speed so the whole functions well.

The difference, of course, is that the aircraft's movements are based on a constant exchange of information and calculations.

It's a different picture of autonomy than the one that dominated public perception a decade earlier, when much of the focus was on the individual self-driving vehicle. Here, autonomy is as much about the ability to coordinate with others as it is about making decisions on one's own.

And the more systems that need to interact, the more important the question becomes of what they're actually optimizing for. Travel time, energy use, cost, battery health, and capacity all have to be continuously weighed against each other.

Much remains

But how smoothly things work here doesn't say much about how it looks in the rest of the world. The Helsinki route is one of the first where the entire autonomous ecosystem actually works together in regular service. Aircraft, airports, charging, energy systems, and air traffic control are integrated well enough that the trip feels effortless.

In many other places, integration between the systems is still a work in progress. Countries have reached different points when it comes to expertise, infrastructure, and implementation, and regulations still differ from country to country. What works seamlessly here may require considerably more human involvement elsewhere.

Education systems have also adapted at different speeds. In several countries, AI, optimization, and autonomous systems have taken up significantly more space in engineering programs, sometimes at the expense of more classical courses. At the same time, a lack of the right skills remains one of the bottlenecks as the technology scales up.

The tip of an iceberg

A while later, we begin our descent toward Helsinki. I look through the presentation I'll soon be giving. The title is: When autonomy became infrastructure.

It's dizzying to think about how much has changed in ten years, and how much still remains.

The flight to Helsinki is, in practice, the tip of an iceberg of calculations. Behind every decision — which altitude, which charging schedule, which route through the airspace — lies an interplay between classical mathematical optimization and AI techniques such as reinforcement learning. The systems have been allowed to test millions of possible scenarios in simulated worlds to learn which choices work best.

What we as travelers experience as a perfectly ordinary flight is the result of a long chain of decisions that constantly need to adapt to one another — in the air, on the ground, and in the energy system.

Maybe that's exactly how autonomy becomes part of the infrastructure. Not through a single spectacular technological leap, but through more and more intelligent systems learning to make decisions together, until the complexity all but disappears from view.

Want to know more?

At RISE, we explore how AI, optimization, and autonomous systems can be used to solve complex problems in areas including mobility, energy, and electrification. Through research and innovation, we investigate how these systems can learn to make better decisions, and how many intelligent systems can interact within an increasingly connected physical world.

Jonas Hellgren is a researcher at RISE, working with reinforcement learning, mathematical optimization, and electrified transport systems, among other things. Together with Johannes Lindgren, he has written the book Reinforcement Learning Explained: A Practical Problem-Solving Approach. The book teaches reinforcement learning in a different way, using many small, concrete examples that build understanding of the methods step by step.

Read more about Jonas's work at RISE.

Read more about the book at relexp.com.