Summary

• Self-driving, autonomous vehicles (AVs) are here. Waymo is testing about 100 cars in London. Uber and Wayve opened a robotaxi waiting list in June. GB-wide AV permits opened in May. The government has projected that up to 40% of new cars sold in the UK could have self-driving capability by 2035.
• For many, AVs are good news. They promise cheaper door-to-door travel, safer roads, independence for people who cannot drive, and a UK market the government expects to be worth around £42bn and 38,000 jobs by 2035.
• Rapid AV adoption will also bring three problems:
• 1. AVs will put hundreds of thousands of livelihoods at risk. England has a record 417,300 licensed taxi and private hire drivers, 120,900 of them in London. Self-driving vehicles will eventually make much of this work obsolete.
  2. AVs will increase congestion. Waymo’s vehicles in California drive 46% of their miles with no passenger on board, and an empty vehicle’s time costs the operator almost nothing. Empty journeys are potentially cheaper than parking, and AVs reduce the opportunity cost of taking journeys, as drivers do not need to focus on the road. The Department for Transport's high-automation scenario predicts a 24% rise in road miles by 2050, with a clear effect on congestion: drivers spend 30% more hours in their cars, average speeds fall by 5%, and motorway speeds fall by nearly a fifth.
  3. Tax receipts will continue to fall. Fuel duty, vehicle excise duty (VED) and the congestion charge currently raise about £33bn a year. As electric cars replace petrol and diesel, the OBR expects fuel duty to fall by about 90% by 2050. AVs are electric, so they pay no fuel duty, and the new electric VED (eVED) replaces only about a quarter of the lost revenue.
• The government needs to act now, before AVs are widely rolled out. There is not yet a substantial constituency of AV owners who will resist a charge; once there is, taxing them becomes far harder. Fuel duty shows the value of acting early: it was introduced in 1909 long before car ownership was widespread, and grew into one of the UK’s largest revenue streams. Had the government waited until car ownership was widespread, introducing it would have been much more challenging.
• This paper sets out the options for how an AV charge could be introduced. We estimate the expected benefit to fiscal headroom in this Parliament at about £0.1bn to £2.8bn a year by 2029-30, with a central estimate of about £0.9bn. 
• By 2050, an AV charge set at the social cost of congestion  (around 88p per mile, depending on time and place) raises £46.6 billion per year – easily covering fuel duty revenue loss, with plenty to spare. This amount is comparable to annual council tax revenue, and is around three quarters of the defence budget. While the revenue will not be realised for years, the benefit can be brought forward as it affects longer dated bond yields today.
• The revenue raised from taxing AVs should be put towards improving public transport, including building metro systems in Britain’s cities and better bus services in towns. Revenue can also be used towards road improvements and new rail infrastructure. Finally, some of the revenue should be used to retrain and support workers who might lose out from the AV transition.

The opportunity of autonomous vehicles

Self-driving, autonomous vehicles (AVs) are no longer science fiction. Waymo is testing about 100 cars in London. Uber and Wayve, a British self-driving technology company, opened a robotaxi waiting list in June this year. GB-wide applications for AV permits opened in May. The government projects that up to 40% of new cars sold in the UK could have self-driving capability by 2035.

The Department for Transport (DfT) projects that AVs will both replace and add to existing miles. Motorists will switch existing, human-driven trips for AV trips. They will also take more trips, through induced demand. On the DfT’s projections, car traffic in 2050 is about 70 billion miles higher than in the central projection, a rise of 24%. Our model points the same way. Without a charge, self-driving vehicles account for 41% of car miles by 2050 and add about 140 billion miles a year to British roads.

These projections are uncertain, and they may be considerable underestimates on both counts: AVs may replace far more existing driving than projected, and they may create far more new travel, and different ways of travelling. For instance, people may choose to work or sleep in AVs, changing the cost of commuting. Conversely, the transition would be slower with taxation, since a charge raises the cost of the marginal self-driving trip. We return to this in the scenarios and trade-offs below.

Getting self-driving vehicles right matters. The potential gains are large, and many of them accrue to the people the current transport system serves worst. The benefits include:

• Cheaper travel, freedom from the cost of a car: Learning to drive and buying a car cost thousands of pounds. After this, running a car costs a household several thousand pounds a year. Robotaxis could allow many households to run one car instead of two, or give up car ownership altogether. This gain is most likely to be felt in the most car-dependent places, where there are fewer public transport options.
• Safer roads: 88% of Britain’s road collisions are caused at least in part by human error. Self-driving vehicles do not speed, drink, get distracted or grow tired. The Society of Motor Manufacturers & Traders (SMMT) estimates that AVs could save about 3,900 lives and prevent 60,000 serious injuries by 2040. The people most impacted by road casualties are pedestrians, cyclists and those in poorer areas.
• Growth and skilled jobs across the country: The government estimates that the UK AV market will be worth around £42bn and create around 38,000 jobs by 2035. The SMMT puts the wider economic benefit at £66bn by 2040. Britain has an early lead, with home-grown firms such as Wayve and Oxa, and a relatively permissive legal framework. The AV charge we propose can be devolved, which would allow regions across the UK to benefit from the transition. AVs could also rejuvenate pubs, whose patrons would have a safe way of drinking and getting home.
• Greater independence for people who cannot drive: Around 3–5% of adults in cities cannot drive because of disability or age. Younger people who cannot yet drive, or who cannot afford lessons and a car, are excluded in the same way. These people depend on expensive taxis, lifts from others, and public transport that is often sparse. AVs offer door-to-door travel on demand, and therefore the freedom to get wherever you need to get: everyday places like work, school, family, friends, hospitals and town centres, but also parts of the country that are harder to get to, and the countryside.

A fair deal for drivers

England has a record 417,300 licensed taxi and private hire drivers. 120,900 of them are in London, and there are around 16,000 licensed vehicles in Greater Manchester. AVs will eventually take over much of this work. There are parallels between the AV transition and the rapid growth in Uber and other private hire vehicles, which has affected cities everywhere. Governments around the world responded to this transition in a variety of ways. In Australia, the governments of New South Wales and Victoria planned ahead and compensated existing taxi drivers with a small charge on private hire journeys. New York, meanwhile, did nothing, and suffered political, social and financial consequences as a result.

• New South Wales legalised Uber in 2015. It funded compensation for existing taxi drivers with a charge of A$1 (53p) per journey, which is negligible enough for passengers to barely notice. The charge raised A$905m (about £470m) of support for existing drivers, up to around $150,000 for each Sydney taxi licence.
• The state government of Victoria, Australia, used a matching charge to fund $100,000 per first licence, plus a hardship fund. The scheme has run for over eight years.
• New York did not offer compensation to existing drivers. Private hire cars led to medallion (licence) values falling from $1.3m to under $200k per licence. In 2018, eight drivers tragically took their own lives. Many drivers bought a medallion on large loans, treating it as their livelihood and pension. In the end the city bailed out bankrupt taxi drivers, costing around $470m.

When it comes to AVs, Singapore opted for retraining taxi drivers. Academies, run by autonomous vehicle operators, train experienced Grab drivers as certified safety and remote operators. 

Learning from these examples, we propose using some of the revenue from an AV charge to compensate and retrain taxi and private hire drivers affected by the AV transition. The revenue can also fund road improvements and other local transport, such as buses, trams and underground lines.

Congestion costs

A further cost from AVs is increased congestion. By definition, self-driving cars lower the marginal cost of driving. Since a human driver is no longer needed, the marginal cost of running an AV is significantly lower: most of the cost of driving a taxi is the labour time of the driver. Likewise, a substantial cost for non-hired vehicles is the opportunity cost of spending that time driving, rather than working, resting or having fun.

Furthermore, the private cost of congestion, for those in AVs, becomes much lower. Time spent in a traffic jam can now be spent working in a comfortable, quiet, air-conditioned office space, or sleeping in a business-class style bed while your car drives itself. But the social cost of congestion remains substantial: people will still need to get from A to B, whether for work, social reasons, deliveries or emergencies. 

These costs will worsen if congestion increases, and if the marginal cost of driving collapses, the number of cars on the road is likely to significantly increase. In cities like London, where congestion is already a challenge, a small increase in road users can lead to disproportionately slower traffic. In some instances, this could be gridlock.

Self-driving vehicles will only deliver the benefits described earlier if they do not also fill our cities with traffic. We outline how this can be mitigated through a charge, which also raises revenue to improve public transport and support those most impacted by the transition to AVs.

The fiscal challenge

Motoring taxes raise about £33bn a year. Most of this is from fuel duty, which brings in just under £25bn. Vehicle excise duty (VED) and the London congestion charge make up the rest. Fuel duty is the 5th-largest source of tax revenue for the Treasury, larger than capital gains, stamp duty or inheritance tax. For decades, fuel duty has funded the British state, but from next year, it is about to drastically fall.

The cause of this collapse is drivers switching from petrol to electric cars. A battery car pays no fuel duty, and (at least initially) it pays a lower rate of VED. As electric cars replace petrol and diesel, the tax base therefore shrinks. The switch is happening quickly, in part due to policy. The government’s zero-emission vehicle mandate requires 80% of new cars to be electric by 2030, and the OBR expects that 90% of all cars on the road will be EVs by 2050.

The OBR expects fuel duty revenue to fall by 90% by 2050, or from around 0.8% of GDP to 0.1%. On the OBR’s own account, this is the single-largest cost of the net zero transition to public finances. Counting VED and other receipts, the total loss reaches £43.2bn a year, 1.0% of GDP, by 2050-51, and keeps rising after that.

The government’s answer so far is the electric VED (eVED), a flat charge of 3p per mile on EVs, which comes into force from April 2028. The eVED helps, but it replaces only about a quarter of the lost fuel duty. This means a substantial gap remains. 

Even if eVED were increased by enough to fully replace fuel duty, this would not solve a more fundamental problem. Both fuel duty and VED are levied on car usage and ownership. This fails to capture one of the most significant costs of driving: congestion. Slower traffic is not only annoying for road users, but economically costly. It is a ‘deadweight loss’, in that no one benefits from the time cost borne by motorists stuck in traffic. 

AVs make both problems worse. They are electric, so they pay no fuel duty. They add traffic, because they create new journeys and drive many of their miles with no passenger. However, AVs also offer a solution. Since they are a new class of vehicle, there is no established group of owners who might resist a new method of taxation. Furthermore, AVs carry a computer that can measure and price its road use precisely, which older cars cannot. This offers a huge opportunity for better tax design. 

To conclude, the government needs a new motoring tax which does three things: 

1. It should provide revenue that grows as fuel duty shrinks.
2. It should price the real cost of road use (congestion and road upkeep).
3. It should be in place before a large group of owners forms to resist it. 

AVs, ostensibly a significant fiscal challenge for the government, are actually well placed to meet all three criteria.

How an AV charge could be introduced

A charge on self-driving vehicles can take several forms. The main design choices are:

1. The charging unit – per minute or per mile
2. The base – does this only apply to empty cars, or any self-driving journey?
3. The rate
4. The level of government at which the charge is administered – UK-wide, devolved regions, mayors or local authorities?

These choices are largely independent of one another, so we can combine different choices to give a range of options. This section sets out the pros and cons of each design choice. The next section models four illustrative combinations.

Some features should hold under any combination. We suggest that the charge should apply only while a vehicle drives itself, i.e. it is in self-driving mode. A person driving manually should not pay the charge, even if their car has self-driving capability. We propose that existing motoring taxes (VED, and fuel duty where applicable) continue to apply to manual driving. This ensures that the new charge addresses the new problems associated with self-driving vehicles, rather than acting as a further charge on drivers. 

For at least the next decade, we expect the charge to fall mostly on commercial fleets of robotaxis, as few households currently have vehicles with full self-driving capability.

Every autonomous journey, or empty only?

The AV charge could apply only to vehicles moving in self-driving mode with no passenger, or to every journey a vehicle makes in self-driving mode. AVs often drive empty between trips, when returning home, while looking for parking, or even instead of parking. 

Driving while empty is, of course, a new problem. In California, Waymos are empty for 46% of their driving miles. In areas with limited or expensive parking, operators may prefer to let their vehicles circulate on the road while they wait for the next passenger. Individual owners may send their cars home, or let them roam, instead of parking them near their workplaces. All these empty miles add to congestion without transporting anyone. 

The main benefit of taxing empty miles is that it targets one of the most novel and negative impacts of AVs, and incentivises operators to ensure their cars are in use as much as possible. It is also cheaper for consumers and less likely to deter adoption. 

However, charging only empty running means the base is smaller, so the charge replaces less of lost fuel duty and forgoes most of the potential long-run revenue, unless it is set at a much higher rate. Non-empty trips still add to congestion, which is not captured by charging empty miles only. A further problem is implementation. It is difficult to verify whether a car is empty or occupied. It is also unclear how authorities should treat delivery or freight. Delivery vehicles are empty of people, but more socially useful than empty passenger cars, and we may not want to tax them at the same rate as empty vehicles. On the other hand, this could lead to tax evasion by reclassifying empty journeys as freight. 

A charge on every self-driving journey makes it easier to price the full congestion cost, raises considerably more, and needs no occupancy rules. On the other hand, it is likely to be much less popular with road users.

Charging unit: per minute or per mile?

Self-driving vehicles will add a large amount of new traffic. They create new demand and, as the Waymo figures above show, half the miles they drive are empty. 

Charging per minute targets traffic more directly than charging per mile. Cars stuck in traffic do not move very far, but impose a substantial time cost. Charging per minute encourages road users to travel at less busy times, combine trips, find parking instead of driving empty, or take different routes to avoid traffic. 

Charging per mile, on the other hand, encourages users to take shorter routes, even if they might take longer. A benefit of charging per mile is that it offers predictability for road users. The time length of a journey depends on traffic, so the final charge is less certain in advance than it would be under a per-mile charge, which is independent of traffic, and depends only on the route. 

The ideal policy would charge per minute, thus penalising congestion, while offering drivers predictability, so they are not hit with charges they did not expect or consent to. There are several ways to achieve this: 

• Rates should be published in advance, so the price of any road at any hour is known before the trip, like an off-peak train fare. 
• For robotaxi journeys, operators can absorb the remaining risk on a portfolio basis, offering riders fixed fares based on the expected cost to them. The same model is used by Uber and other private hire operators today, which offer customers a guaranteed price even as costs vary per journey. 
• Private users facing unexpected congestion would retain the option to switch the vehicle to manual driving, in which case the charge stops. 
• If authorities wished to reduce unpredictability even more, they could cap the charge per journey, though this reduces the congestion disincentive.

A per-mile charge is simpler and more predictable, and it matches the design of the eVED, which may help with administrative simplicity. But to properly target congestion, a per-minute charge creates better incentives, and we believe that implementation hurdles can be overcome.

What rate? 

The rate of the charge could be set according to different policy objectives. Two obvious candidates are (1) a rate equivalent to what a petrol or diesel car pays per mile, or (2) a rate that covers the full social cost of congestion caused by AVs. 

(2) is a much larger sum than (1), so replacing the per mile cost of fuel duty keeps the charge lower for users, and leads to a faster rollout. The downside is that (1) only captures some of the social cost of AVs, as the rate is too low to adequately discourage and compensate for AV-induced congestion.

On an empty-running base, no rate up to the social cost replaces the revenue generated by fuel duty.

Setting the rate to cover the full social cost of congestion reflects the cost each vehicle imposes on others by using scarce road space at a busy time. We calculate that this cost would be around 88p per mile on the roads self-driving vehicles are likely to use in 2030. 

Even at the higher rate, most journeys stay cheap, because the charge is concentrated where and when road space is most in demand. Indeed, the charge could be near zero during off-peak periods and on clear roads. 

These two benchmarks are not the only options, or indeed a binary choice. The government may want to consider setting rates at the lower benchmark first and setting a path towards the full social cost, or providing mayoral authorities the option to set rates higher, as adoption and the evidence base grow.

National or devolved?

We suggest that whether the charge is set and applied locally or nationally, there should be one national system which works across the country. We do not want drivers to have to sign up to different websites or payment methods depending on where they drive. However, even within a single national system, there is room to empower mayors (or local authorities) to set different rates and retain revenue locally.

There are several options.

• UK-wide. The Westminster government sets rates and scope for the whole of the UK. Revenue goes to the Exchequer (but could be distributed to regions depending on where revenue is raised).
• Devolved to nations and regions. Parliament lets Scotland, Wales, Northern Ireland and English combined authorities set their own rates on their own roads, zone by zone, and keep what they raise for local transport and driver compensation. Mayors already decide whether these fleets can operate. Permits for self-driving passenger services need the local transport authority’s consent. Letting them set the price is consistent with this devolved power. A benefit of this is that it would give English mayors their first genuinely new and growing tax base, distinct from grants or shares of national taxes. Providing this option could also allow areas that wish to move faster on approving AVs to do so, as there is an additional fiscal benefit. 
• Combined. A UK-wide floor, but devolved regions are free to add additional rates above it and keep the difference. 

Whichever model is adopted, Parliament should establish an independent national body to administer the charge. This would ensure a single technical standard and set any rates that are not devolved, updating them as often as is practicable. 

Under any option, local authorities retain powers to keep certain classes of vehicle off certain roads. For instance, authorities in rural areas should be able to keep self-driving lorries off minor roads.

Modelled design choices

We model four illustrative combinations of the choices above. Each design is a combination of three choices: what gets charged (every car mile / self-driving mile only / empty mile only), what rate (full or duty), and what unit (per mile or per minute). The two rate options are:

1. Full rate = designed to cover the congestion cost of driving on a road, based on the Department for Transport's Transport Analysis Guidance calculation of the marginal external cost depending on the zone and time. This is not a single number: it varies from about 9p per mile on a motorway or an overnight road to about £2.62 per mile at the London peak, or from 5p to 45p per minute. The zone table above gives the full set. Averaged across the roads self-driving vehicles are likely to use in 2030, it comes to about 88p per mile. 
2. Duty rate = the fuel duty an equivalent petrol or diesel car pays per mile on that road, converted at the zone's speed where the unit is per minute. It's roughly 5–6p a mile for a typical petrol car (the OBR describes the eVED's 3p as ‘around half the fuel duty rate’). So a duty-rate charge simply replaces the fuel duty that electrification is losing, like for like — it doesn't price congestion at all.

Revenue (£bn)

Percentage of fuel duty revenue covered

Designs 1 and 2, which tax empty miles only, do not raise enough to replace fuel duty revenue.

However charging empty self-driving miles at the full congestion cost raises £0.2bn in 2030 and £12.2bn in 2050, against the £26.1bn and £34.9bn needed to hold fuel duty's share of GDP. 

Coverage rises from under 1% to about 35%. Charging per minute instead of per mile raises slightly more, £13.4bn by 2050, or 38%. Neither closes the gap, because empty running is a small share of traffic. Only a charge on every self-driving journey grows into a substantial revenue source, and only once self-driving take-up is high.

Impact on fiscal headroom

How much an AV charge raises, and how much fiscal headroom it creates, depends on the combination of choices set out above. 

Any credible charge, however, has an immediate benefit for public finances, even if revenue does not substantially increase for years to come. This is for two reasons:

1. Lower long-term gilt yields

The OBR bases its debt-interest forecast on market yields. So if a credible long-term revenue stream lowers gilt yields (the interest rate on UK government bonds), debt interest falls within the forecast period, increasing immediate fiscal headroom.

Markets price in the whole future revenue stream, which (according to our forecasts) reach £89 billion when every car is an AV, and £46 billion by 2050, under the fuller versions of the charge. Investors’ doubts about UK public finances show up most in long-dated gilt yields, and those yields respond to today’s policy decisions. In autumn 2025, analysts put the extra yield investors demand for that risk at about 25 basis points (bp) on 10-year gilts. On Budget day in November that year, a larger-than-expected consolidation cut 30-year yields by about 10bp within hours. A charge that closes the largest known gap in the long-run tax base, fuel duty, is likely to be treated similarly.

The impact on public finances is immediate: on the OBR’s ready reckoner, each 10bp off gilt yields is worth about £0.84bn a year by 2029-30.

2. The OBR five-year forecast

The OBR forecasts five years ahead, currently to 2030-31. It judges the Chancellor’s headroom against the current-budget rule in 2029-30. Any charge that raises money inside that window counts and adds to headroom. Receipts depend on how much self-driving mileage there is to charge by 2030-31. On any realistic path, this is a small share, likely a few thousand to a few tens of thousands of robotaxis by 2031, under 1% of national mileage.

The eVED shows how this can create immediate headroom. The government announced eVED in the 2025 Budget, to start in April 2028. The OBR scored it at £1.1bn in 2028-29, rising to £1.9bn in 2030-31, with £1.4bn in the 2029-30 rule year, because those years fall inside the forecast. A charge on self-driving vehicles that started alongside the eVED in April 2028 would score the same way. The forecast also moves forward a year at each Budget so, over time, more of the revenue falls inside the forecast window.

This charge also helps the second fiscal rule, which requires net financial debt to fall as a share of GDP by 2029-30. Since the gilt effect saves money on debt interest, the debt stock would be lower than the counterfactual by 2030. Cheaper gilts also make any given investment cheaper to fund, which widens the room to borrow for capital projects within the rules (we do not include this particular effect in our model).

Finally, unlike many taxes, an AV charge could raise GDP. This is because it is taxing a negative externality; namely, congestion, which poses a deadweight loss to the economy. If AV taxation allows local areas to permit AV adoption faster, this could also lead to benefits for the growing British AV industry. 

Implementation

Under the Automated Vehicles Act 2024, a vehicle must be authorised by the Department for Transport as self-driving before going on the road. This means it is to meet the self-driving test, whereby the car is deemed to be at least as safe as a careful, competent human driver. An Authorised Self-Driving Entity (ASDE), typically the manufacturer or provider of the self-driving technology, must take legal responsibility for how the vehicle drives. 

We propose that paying the charge and running a compliant on-board system should be a condition for operation in the UK, and that the Automated Vehicles Act should be amended to require this. The Secretary of State can suspend or withdraw authorisation.  A failure to pay, or to run a compliant on-board system, would be both a tax matter and a breach of the amended Automated Vehicles Act. 

How can we protect privacy? 

Each autonomous vehicle would have software to work out the charge, based on location and time spent in self-driving mode (or based on mileage, if using a per-mile charge). Only the sum owed, the vehicle ID and anonymised area breakdown would be shared with authorities. The law should set criminal penalties for misuse. The government should involve civil liberties groups in the design from the start.

How does the charge work with the eVED? 

VED stays as it is, a registration tax on all vehicles. The on-board system records the split between human-driven and self-driving miles. Human-driven miles go into the yearly eVED return as normal. Self-driving miles are covered by the new AV charge. Annual MOT checks can ensure that the totals add up against the odometer.

How should we treat HGVs? 

A per-minute charge rewards fewer, fuller vehicles. A large lorry uses far fewer vehicle-minutes per tonne than several vans, so it pays less to move the same goods. However, while larger vehicles impose less congestion cost, they create more road damage, as they are heavier. 

If authorities are concerned about the tax burden on HGVs, one option is to replace the HGV levy with the AV charge for self-driving lorries. Alternatively, they could keep both rates for HGVs on non-A roads, in order to keep self-driving freight on the main road network and stop lorries diverting through villages.

Trade-offs and objections

Would the charge slow the rollout of self-driving vehicles? 

Most likely, yes. A charge increases the cost of adoption and therefore slows the transition. To some extent, this is desirable: it gives workers and cities more time to adapt to the AV revolution, while raising money to help pay for retraining, compensation and investment in public transport. 

A rapid transition could be damaging, not just for private hire drivers, but for all road-users, who could face sudden gridlock. A charge set at the cost of congestion deters the trips worth less than the delay they cause others. The charge will be near zero off-peak and on clear roads, so it barely touches most of the rollout: suburban, inter-urban and overnight travel. The effect falls on busy city centres at busy times, which is where we want it.

That said, a charge could speed up AV rollout in areas that are keen to take advantage of the fiscal benefits produced by the AV charge. As things stand, regions may be reluctant to permit robotaxi operators in their cities because of the impact on congestion and workers. An AV charge offers a way to resolve this trade-off, and could lead to faster rollout in some areas.

Would it raise prices for road-users? 

While a charge increases the cost of a journey, it significantly reduces the cost of congestion. This is a net benefit for road users. Congestion is a real cost, imposing an economic cost of around £85bn annually. This charge aims to eliminate that cost. 

Furthermore, the revenue generated from the AV charge is not lost: it can be spent on road improvements and public transport, which could reduce the cost of living through cheaper rail, bus and metro fares. If road improvements can be funded through the AV charge, this may also reduce the tax burden in other areas. 

Of course, the inevitable effect of a charge is that it will raise the cash cost of some journeys, while reducing the time cost. Pricing road space will change behaviour. Some journeys move to buses, trams or quieter times, which cuts congestion. Some costs, such as autonomous grocery deliveries, get passed on. Our figures already include the fall in demand, through two channels: (1) private drivers can switch to manual driving to avoid the charge, and (2) robotaxi demand responds to the charge through fares. What they do not model is who finally bears the cost; i.e. how much is borne by operators, or passed on to passengers and other consumers.

Is the charge too complex to understand? 

Implementing an AV charge, especially one that is sensitive to time and location, is complex. However, the complexity sits with the rate-setter, not the traveller. The rate-setter publishes rates in advance, so a rider sees the fare before the trip, like an off-peak train fare or a quoted taxi fare. This means there is no unexpected tax bill. For at least the next decade, we expect that the charge will mostly fall on commercial fleets in self-driving mode, rather than households with their own AVs.

Distributional impact

The distributional effects of an AV charge depend heavily on the model adopted. A low-rate charge on empty AVs only has almost no direct incidence on passengers, while a charge on every journey at the full social cost shifts costs towards peak-time city-centre trips. 

Our analysis below is based on the fullest version, a per-mile charge on every self driving mile at the social cost of congestion. Less radical versions of an AV charge lead to more modest impacts.

Congestion is itself a type of charge, paid in time rather than money. Since no one collects the proceeds, it is a ‘deadweight’ cost. Being stuck in traffic is disproportionately bad for those who can least afford to wait. This is likely to be people whose work depends most on being in a physical space, who cannot work from the back of an AV. Jobs which require in-person interaction, such as social care, education or hospitality, are likely to be most affected, along with those in delivery and logistics, which require moving goods from A to B. 

Many of the costs will be borne by those who would not pay an AV charge. For instance, buses cannot easily route around traffic, so they absorb delay in full unless road space is specifically reserved for them. Around 40% of households in the lowest quintile have no car at all, compared to 14% in the highest. These households already pay for congestion in slower journeys, but receive nothing in return.

The proposed charge is highest where road space is most in demand. This means that passengers in wealthier, urban centres, travelling at peak times, are most affected. Under the full social-cost option, the model assumes a peak rate of about 262 per mile (about 45p per minute) in central London, and close to zero on uncongested roads outside peak hours. Meanwhile, for off peak hours in a small town, the charge is close to zero. Where a self-driving vehicle carries several passengers, the charge is divided across them. The per-rider cost is a fraction of that for a single-occupancy robotaxi, so the modes used most by lower-income households are the least affected.

While these factors imply that an AV charge falls most on the most well off, it is worth noting that flexibility is unequally distributed. A shift worker with a fixed start time, or a parent tied to the school day, may struggle to travel off-peak to avoid the charge. That said, it is precisely these travellers who benefit most from reduced congestion, which is the primary target of the AV charge. 

Overall, the distributional outcome ultimately depends on what the revenue is used for. Spent on buses, trams and local transport, and on compensating and retraining the drivers displaced by the transition, the charge is efficient, growth-enabling, and highly progressive.

With thanks to Ben Southwood, Pedro Serôdio and John Myers for helping to design the proposal, and to James Howat, Joe Allen, Dan Mead, Michael Dnes, Tym Syrytczyk, Peter Roberts, Michelle Gordon and James Hutt for providing feedback.