Talking to the Machine

The old racing-driver habit of talking to the car - assuming such a habit has survived our digital age at all - is gradually acquiring an entirely new meaning. The pity is that the car still does not understand the driver, while the driver is increasingly unsure that he understands the language spoken by the thing whose steering wheel he has spent years turning with relative confidence. And, more to the point, he has no idea what the thing intends to do in the next corner.

At the Belgian Grand Prix, Oscar Piastri ran into a rather peculiar problem. Through the corners his lap was practically identical to Lando Norris’s, the power-unit settings on the two McLarens were almost the same, yet on the straights Piastri was losing time. By the end of the lap the deficit had grown to roughly two tenths of a second.

Andrea Stella’s explanation was even more inviting to those of us inclined to suspect supernatural interference. The difference came from “self-learning elements” in the power-unit control system. The computer constantly adjusts how electrical energy is deployed according to how the car negotiated previous sections and previous laps. So the same throttle position and nominally the same selected mode no longer guarantee that the driver receives exactly the same combined power at exactly the same moment.

Piastri himself phrased it rather less academically: when grid position begins to depend on whether the computer happened to behave correctly, this is not his favourite way of going motor racing.

Quite. But what exactly is the computer doing - and why did Formula 1 reach the point where a sizeable chunk of the driver’s job had to be handed over to it in the first place?

Look What We’ve Come To

Before 2026, the MGU-K could deliver a maximum of 120 kW. The limit is now 350 kW, while the internal-combustion engine has been reduced to roughly 400 kW. In other words, the electric side has ceased to be a relatively modest supplement to the petrol engine and now supplies almost half the power of the entire power unit.

We shall spare you, for perhaps the hundredth time, the explanation that a substantial share of the energy required by this battery does not come elegantly sailing back into it under braking. Some of it is obtained rather more directly from the internal-combustion engine, which is forced - at some of the most famous places on some of the world’s most famous circuits - to spend part of its working day as a generator. At which point television directors, lowering their eyes in embarrassment, tend to abandon the onboard camera for something shot from a respectful distance.

As a technical toy, all of this is fascinating. As the compulsory and only acceptable way forward, forgive us for remaining considerably less enchanted. The important point is that you cannot simply spend your beloved electricity continuously. Not, at least, until somebody finally packages a tokamak behind the driver’s seat!

First, the energy has to be recovered through the MGU-K: under braking, with the throttle partly closed, on lift-off, or during the so-called super-clipping phase. In the last case the driver may still have the throttle pedal fully depressed, yet the control electronics reduce tractive output and force the MGU-K to generate electricity instead. The car begins slowing before the conventional braking point, but stores energy to be used later.

If this logic has not yet driven you into a mild stupor, we may continue. Unfortunately, it gets worse.

Most of these processes are automatic. Formula 1 explicitly states that braking regeneration, partial-throttle regeneration and super-clipping are controlled by the electronic control unit according to selected maps and predefined targets. The driver can directly provoke regeneration by lifting the throttle, but much of the rest belongs to the machine.

Suppose there are three acceleration zones ahead. There is not enough stored energy to provide the full 350 kW all the way through all three. Where do you spend it? Immediately after the slow corner, where extra power cuts lap time most efficiently, or at the end of the first straight? Should regeneration begin slightly earlier before braking, sacrificing a few kilometres per hour now in exchange for more power several seconds later?

For one perfectly fixed lap, such a problem can be optimised in advance. The slight difficulty is that a real lap is never perfectly fixed.

Why the System Has to “Learn”

The driver takes a corner 2 km/h faster. He opens the throttle earlier, brakes slightly later. A headwind appears on the straight, grip changes. Every one of these apparently tiny deviations changes the future energy balance.

If the car spends longer accelerating at full throttle, the original deployment plan may empty the battery before the point at which it was meant to. If the driver lifts earlier, an additional opportunity for recovery suddenly appears.

So the system cannot simply replay a rigid instruction: 350 kW here - 200 there - charge the battery here. According to Stella, the algorithm looks ahead.

Using the actual data from the lap, it recalculates how the remaining electrical energy should best be used. And these corrections take place both from lap to lap and within the lap itself. This is what journalists have been calling “self-learning”.

One should be a little careful with the phrase. There is no need to imagine some miniature Google AI sitting inside the McLaren, pondering its own strategy for the power unit - although, judging by the quality of some AI “thought processes”, we admit we would quite like to watch that experiment.

What we are actually dealing with is a predictive and adaptive control algorithm. The model receives new input data, compares reality with its expected state and adjusts subsequent energy deployment. The principle is familiar from control theory: instead of merely reacting to what is happening now, the system chooses an action intended to produce the best result several seconds into the future.

Very clever. But we remain stubborn enough to think that a racing driver might still like to know what his racing car is about to do.

Why Even the Out-Lap Matters

Piastri pointed to another unpleasant feature: what the driver does on the out-lap can influence the energy strategy of the following flying lap. From an engineering standpoint this is perfectly logical.

The car does not begin the next lap in some idealised abstract state. The battery state of charge is what it actually is, temperatures are what they actually are, and previous regeneration has already happened. The system has already observed the driver’s acceleration and braking behaviour, the actual timing through different sections is already known.

So two drivers may select the same nominal mode yet arrive at the beginning of the flying lap with slightly different initial conditions - and the system may then construct different energy trajectories for them.

In the old hybrid formula, a small difference in deployment represented a relatively small part of the overall result. At 350 kW, it can decide the lap.

Everything would be marvellous if the cockpit contained a Terminator. It does not.

How a Driver “Breaks” the System

Isack Hadjar provided perhaps the clearest example at Spa. During qualifying he wanted to give Max Verstappen a tow, so after Turn 14 he slowed in an unusual way and waited for his team-mate. To a human being, the situation could scarcely have been simpler. Hadjar was waiting for another car.

To the algorithm, something had happened which had absolutely no business happening on the lap it had expected. Actual speed, throttle position and section time suddenly diverged sharply from the prediction. The electronics then recalculated the energy strategy for what remained of the lap.

The outcome was almost comic. On one attempt, after Hadjar returned to the throttle, the car gave him more power than he expected and he pulled away from Verstappen too quickly. On the next, there was not enough energy available and Verstappen began catching him instead.

The system had not “gone mad”. It was doing its best to solve the mathematical optimisation problem after receiving unusual input data. The problem was that the driver could no longer confidently predict the answer to that mathematical problem.

And at 300 km/h, “I wonder what the optimiser has decided this time” is perhaps not the most reassuring thought to carry into the next braking zone.

The Wind Gets Involved in the Engine

Andrea Stella offered an even more revealing example involving a headwind. Suppose the car enters a straight with a stronger headwind than the one assumed during simulation. It accelerates more slowly and therefore spends longer on that section of track. That changes the duration for which electric assistance is required.

The algorithm observes what is actually happening and adjusts the deployment plan. From its point of view, this is entirely rational. From the driver’s point of view, the result may look very different: at the point on the straight where the power unit pulled strongly yesterday, electric assistance may begin fading earlier today.

This is why Stella includes not only the driver’s own actions among the important disturbances to the model, but also wind and available grip - quantities which cannot be represented perfectly in advance.

So we have reached the stage at which a gust of wind does not merely change the aerodynamic drag of the car. It also gets a vote in what the engine is going to do next. Progress is a wonderful thing.

Where Has My Braking Point Gone?

Stella gave another useful example. If the system decides to use stronger super-clipping before a corner, the car may arrive at the normal braking zone roughly 10 km/h slower.

Now imagine a driver who has spent dozens of laps braking at the same marker. On one lap the car arrives there at 320 km/h. On another, because the recovery strategy has changed, it arrives at 310. The downforce, the amount of kinetic energy to be removed, the initial load transfer – everything is slightly different. Slightly but noticeable for a professional driver.

Therefore the old braking point is no longer optimal. So the electronics are not merely changing the maximum speed on the straight. They are indirectly changing the driver’s physical reference points for controlling the car.

And this, perhaps, is the real source of the drivers’ irritation. They are accustomed to adjusting their driving for grip, wind and tyre condition. But now another rapidly changing layer sits between the throttle pedal and the car’s response: the energy-management strategy.

A racing driver can adapt to a car that is difficult. He can adapt to one that is slow. He can even adapt to one that is trying rather enthusiastically to kill him - motorsport history is quite well supplied with examples. What is considerably harder is adapting to a car whose answer to the same question is not always the same.

The Russell Case

Not every strange behaviour seen in 2026 should be blamed on the adaptive algorithm.

George Russell spent several rounds complaining that his Mercedes was unexpectedly losing electrical drive on the straights compared with Andrea Kimi Antonelli’s car.

Initially, the team suspected driving style. Russell even began changing the way he used the throttle in an attempt to copy his team-mate. At Spa, the problem remained.

Further investigation revealed an actual software defect. The electronics were deploying too much electrical energy too early in the lap. Anything spent unnecessarily in the opening sections was, naturally enough, absent later when Russell expected maximum assistance from the MGU-K.

This was a different problem. Piastri was dealing mainly with the extreme sensitivity of a correctly functioning algorithm to changing conditions. Russell had an error in the software itself.

For the driver, however, the cockpit sensation could be remarkably similar: the throttle is fully open, and the car does not accelerate the way you expected it to.

Why Not Give the Driver a Great Big Red Button?

The obvious solution is to remove all this automation and hand the 350 kW to the driver. Press the button if you want it. Do not press it if you want to save the energy. Simply lovely.

On 6 August, this very question was put directly to FIA Single-Seater Technical Director Nikolas Tombazis. His answer was instructive.

If the driver were simply given an on/off control, the rational strategy would be obvious: deploy maximum electrical power on corner exit, where it gives the greatest reduction in lap time. The trouble is that the energy reserve would disappear very quickly.

The car would run out of electrical assistance even earlier on the straight and the unpleasant fall in combined power - clipping - would become worse.

At the same time, the driver would have to respect limits on regeneration, battery state-of-charge variation, MGU-K power and energy allocation in different parts of the circuit. According to Tombazis, a human being simply cannot manage all those parameters rationally at once.

And this is where the real problem lies. The problem is not that an algorithm exists. The problem is how much authority has to be handed to that algorithm because of the underlying energy concept of the power unit.

If you design a system which only works properly when a computer continuously predicts, juggles and conceals its shortages from the person driving the car, then perhaps the computer is not the part of the concept worth questioning first.

A Living Issue

The electronics in a 2026 Formula 1 car are doing exactly what a good automatic control system is supposed to do. They collect large amounts of data, estimate the system state, predict future demand and adapt to changing conditions (in the way they “think” should be better for the whole humanity).

From a purely engineering perspective, this is a genuinely fascinating problem. But a racing car contains one more participant in its control loop.

A human being.

And if the car’s internal optimiser is so sensitive that a few metres’ difference in throttle application, a gust of wind or an unusual out-lap can alter the subsequent behaviour of the power unit, the driver begins to lose the most important thing required when operating a machine at the limit: a clear causal link between what he does and what the car does in response.

A modern hybrid Formula 1 car really does require sophisticated automatic energy management. It cannot simply be deleted with one regulation change. The FIA openly acknowledges that. In fact, removing it without changing the power-unit concept could make the underlying energy shortage even more obvious.

And let us be clear: we have no objection to electrical systems as engineering toys. Give engineers another motor-generator, a battery, four inverters and a month in the workshop and we shall happily spend the evening examining what they have invented.

What grates is the moment when an interesting technological option is promoted into the only permissible future, and racing itself is then bent around making that decision function.

By now it has become obvious that the great electrification experiment reached far beyond merely changing the source of part of the power.

It changed braking, throttle response. It even changed what a braking marker means from one lap to the next in a non-predictable (at least at human brains speed) way.

And after all that, people finally arrived at a conclusion which could perhaps have been written on the first page of the project twenty years ago: A good algorithm must do more than spend energy in the mathematically optimal way. Its behaviour must remain predictable enough for the human being controlling the car at 300 km/h.

At Spa, Formula 1 discovered this with unusual clarity - although, to be fair, there were people warning them from the beginning that the whole idea was becoming rather absurd.

For the coming seasons, the movement is already beginning to reverse. The FIA, teams and manufacturers have agreed to gradually alter the balance between the internal-combustion engine and the electrical side: the ICE is set to regain some of its share, while the result should become less dependent on elaborate energy management.

The problem will not disappear completely. But even Tombazis now admits that the present system has become more complicated than anyone would ideally like.

Which is usually how these things end. First we invent an extraordinarily “clever” machine to solve a problem. Then we spend several seasons trying to make the machine slightly less clever so that the racer can drive the flipping car again.

• < Назад
• Вперёд >