Quantum Annealing Tested for Real-Time Micro-Mobility Fleet Dispatch

Urban transportation systems increasingly depend on the ability to position vehicles where customers are most likely to need them.

For conventional taxi or ride-hailing operations, this can involve deciding which vehicle should serve which passenger.

For shared electric vehicles and other forms of micro-mobility, the problem can be different.

A fleet operator may need to decide where vacant vehicles should wait, which vehicles should respond to new requests, and where vehicles should be positioned for future demand.

These decisions can change continuously as customer requests arrive.

A peer-reviewed study in Scientific Reports examined whether quantum annealing could be used to address this type of dispatch problem. The paper, by Takeru Goto and Masayuki Ohzeki, presents a new formulation for micro-mobility dispatch that incorporates both historical demand information and current vehicle positions. The article's version of record was published on August 5, 2026.

The research is relevant to transportation because it focuses on a practical fleet-management problem rather than a purely abstract optimization exercise.

At the same time, the results should be interpreted carefully.

The researchers tested the approach in simulations and on a D-Wave Advantage 1.1 system. The work does not represent a commercial deployment of quantum dispatch technology.

Instead, it provides an experimental evaluation of how quantum annealing could be incorporated into a transportation optimization workflow.

The problem with conventional route planning

Many transportation optimization systems are based on variants of the Vehicle Routing Problem, or VRP.

VRP models can be used to determine how vehicles should be assigned to customers and how routes should be organized.

However, the researchers argue that conventional VRP formulations are not necessarily ideal for micro-mobility dispatch.

The reason is that micro-mobility fleets can operate in highly dynamic environments.

Customer requests arrive continuously.

Vehicles become available at different locations.

Vehicles may need to return to charging or waiting stations.

The location of future demand can also vary substantially throughout an operating period.

In such an environment, planning a long sequence of vehicle stops may be less useful than continuously deciding where available vehicles should be positioned.

The study therefore develops a separate formulation called the Micro-Mobility Dispatch Problem, or MMDP.

Why historical demand matters

A key feature of the research is its use of historical usage information.

Historical data can show where customer requests are more likely to occur and where customers are likely to travel.

The researchers incorporate this information into the optimization model using a Bayesian approach.

The resulting model estimates how many vacant vehicles should ideally be allocated to particular stations.

The purpose is straightforward.

If a station is located near an area that historically generates many requests, positioning vehicles there before demand arrives may reduce the time required to respond to customers.

The optimization therefore considers both current vehicle positions and historical demand patterns.

This creates a link between data-driven forecasting and optimization.

Rather than simply sending the nearest vehicle to the next request, the system attempts to position the broader fleet in anticipation of future demand.

How the quantum model works

The researchers convert the dispatch problem into a Quadratic Unconstrained Binary Optimization, or QUBO, formulation.

QUBO models use binary decision variables to represent possible choices.

In this study, variables indicate whether a vehicle is assigned to a particular customer or station.

The model includes constraints ensuring that vehicles receive appropriate assignments and that cu

stomers are assigned to vehicles.

It also incorporates travel time and the desired distribution of vehicles among stations.

This formulation allows the problem to be processed using quantum annealing.

Quantum annealing is a quantum-computing approach designed for certain optimization problems that can be expressed in forms such as QUBO.

The researchers used this formulation both in their simulation experiments and in tests on D-Wave quantum hardware.

Dynamic and static dispatch

The study evaluates two versions of its proposed approach.

The dynamic approach uses current vehicle positions when determining how vehicles should be distributed.

The static approach relies more heavily on statistical information and does not continuously adjust station allocation based on every vehicle's current position.

The distinction creates an important operational trade-off.

The dynamic approach can react more directly to the current state of the fleet.

But that responsiveness can also cause vehicles to move more frequently between stations.

The static approach is less reactive, but it can avoid some unnecessary repositioning.

The researchers found that the dynamic approach achieved the strongest customer-service performance, particularly in terms of customer waiting time.

However, this improvement came with increased total vehicle travel because vehicles were actively repositioned according to changing conditions.

The static approach provided a more balanced result, improving service quality without a significant increase in travel distance.

This trade-off is important for transportation operators.

Reducing customer waiting time is valuable, but additional vehicle movement also consumes energy and can increase operating costs.

A dispatch system therefore cannot optimize only one metric.

What happened in the simulations?

The researchers first evaluated the proposed formulation in a simulated transportation environment.

Their default simulation used a grid-based environment containing six vehicles.

Customer requests were generated according to a statistical model, with higher-demand areas receiving more frequent requests.

The researchers compared the proposed approaches with a distance-based greedy strategy and a VRP-based QUBO formulation adapted to the micro-mobility problem.

The results showed that the proposed dynamic and static formulations outperformed the baseline approaches on the study's customer-service metrics.

The dynamic formulation produced the lowest customer waiting times.

However, the researchers also observed that the dynamic system increased total travel because of the additional repositioning activity.

This is an important result because it demonstrates that optimization objectives can conflict.

A system designed to minimize waiting time can potentially increase vehicle movement.

A system designed primarily to minimize movement may leave vehicles poorly positioned for future customer requests.

The appropriate balance depends on the operational objectives of the transportation service.

Demand concentration affects the results

The researchers also tested how the approach behaved when customer demand became more geographically concentrated.

This is relevant to urban mobility because customer requests are rarely distributed perfectly evenly across a city.

Some areas can generate much higher demand than others.

The study found that the proposed methods were particularly effective when demand was spatially concentrated.

The advantage became more pronounced as the demand distribution became more uneven.

The researchers also found that when demand was relatively uniform, the simple greedy strategy could perform better than the proposed approach.

That finding is useful because it demonstrates that quantum-inspired optimization is not automatically superior under every operating condition.

If there is little useful structure in historical demand, the additional complexity of the proposed model may not provide an advantage.

This reinforces the importance of matching an optimization method to the characteristics of the operational problem.

Testing on real quantum hardware

The researchers did not stop with simulations.

They implemented the proposed formulation on a D-Wave Advantage 1.1 quantum annealing system.

For comparison, they used Gurobi Optimizer 13.0 on a conventional computer equipped with an AMD Ryzen Threadripper PRO 5995WX processor and 256 GB of RAM.

The comparison used time-to-solution, or TTS, as one of the performance measures.

TTS estimates the time required to obtain an exact solution with a specified probability.

The researchers also considered both the pure quantum annealing time and the broader quantum-processing access time.

This distinction is important.

A quantum processor may perform the actual annealing operation extremely quickly, but the complete process can include programming, data loading, readout, and other access delays.

The researchers specifically noted that QPU access time was significantly longer than the pure annealing time in their experiment.

What did the hardware test show?

The quantum hardware results were more limited than the simulation results.

Using reverse annealing, the researchers attempted to solve 200 problem instances.

The method reached exact optimal solutions for only a subset of those instances.

For the static formulation, 13 instances were solved exactly.

For the dynamic formulation, 28 instances were solved exactly.

For those instances that were solved exactly, reverse annealing demonstrated better time-to-solution than Gurobi in some cases, including when the researchers considered total QPU access time.

However, performance varied substantially between instances.

This is a critical qualification.

The results do not show that quantum annealing is generally faster than classical optimization for micro-mobility dispatch.

They show that the quantum method achieved favorable results on a subset of the tested problems under the study's experimental conditions.

That is a much narrower—and more defensible—finding.

Quantum annealing versus simulated quantum annealing

The researchers also compared actual quantum annealing with Simulated Quantum Annealing, a classical computational method designed to emulate aspects of quantum annealing.

The comparison produced an interesting result.

For forward annealing, the quantum approach performed better.

For reverse annealing, however, simulated quantum annealing performed slightly better under the tested conditions.

This result demonstrates why comparisons with classical methods remain important.

A quantum system producing a good solution does not automatically mean that quantum hardware is providing a computational advantage.

A classical approximation of the same approach can sometimes perform similarly or better.

For logistics companies, this is ultimately what matters.

The useful question is not whether a system uses quantum hardware.

The useful question is whether the complete system produces better operational results at an acceptable cost and speed.

Scalability remains an open question

The researchers also examined how the proposed approach behaved as the fleet size increased.

The experiments showed that performance deteriorated when solver parameters were held constant as the problem size increased.

Scaling the number of reads helped prevent this degradation and allowed the proposed approach to maintain its advantage over the greedy baseline in the tested scenarios.

However, the authors caution that the probability of reaching the exact optimal solution decreases as problem size grows.

They state that calculating exact time-to-solution for real-world fleet sizes would be practically prohibitive at the current experimental scale.

This is one of the most important limitations for commercial transportation.

A research model working with a relatively small fleet is not automatically ready for a city-scale transportation network.

Large commercial fleets can involve many vehicles, stations, requests, charging constraints, traffic conditions, and other operational variables.

Those conditions need to be tested separately.

Why this matters for logistics

Although the study focuses on micro-mobility, its broader relevance extends to fleet management.

The underlying problem is one that appears across transportation:

Where should available vehicles be positioned before the next demand arrives?

A similar question can arise in delivery fleets, shared vehicles, service vehicles, and other on-demand transportation systems.

The answer can depend on historical demand, current fleet locations, travel times, and operational constraints.

Optimization systems that can process these variables quickly could potentially help transportation operators respond more efficiently.

The study does not demonstrate these broader applications.

But it provides a research framework that could be investigated in other fleet-dispatch environments.

What businesses should watch next

Transportation companies interested in quantum optimization should watch several areas.

Larger fleet experiments

Future studies need to determine whether quantum-assisted dispatch remains useful as the number of vehicles and stations increases.

Real transportation data

The study uses simulated environments.

Testing with larger real-world datasets would provide stronger evidence about practical performance.

Energy and operating costs

Reducing customer waiting time is only one objective.

Additional vehicle repositioning can increase travel and energy consumption.

Future research should evaluate these trade-offs together.

Full system latency

Quantum processing time is only one part of an operational dispatch system.

Data collection, model construction, QPU access, result processing, and dispatch execution all contribute to total response time.

Strong classical alternatives

Quantum approaches should continue to be compared with modern classical optimization methods and heuristics.

A quantum system needs to demonstrate value against the technologies businesses can already deploy.

Conclusion

The August 5, 2026 version-of-record publication of research into quantum-annealing-based micro-mobility dispatch provides another concrete example of quantum computing being tested against a transportation optimization problem.

The researchers developed a QUBO formulation that combines historical customer-demand information with current vehicle positions to determine how vacant vehicles should be assigned to customers or positioned at stations. The method was evaluated through simulations and on D-Wave Advantage 1.1 quantum hardware.

The simulation results were encouraging within the study's defined environment. Both the dynamic and static approaches outperformed the selected baseline methods on customer-service measures, while the dynamic approach achieved the lowest customer waiting times. However, its stronger responsiveness also increased total vehicle travel.

The hardware results were more limited. Reverse annealing found exact optimal solutions for only a subset of the tested instances, although it achieved better time-to-solution than Gurobi for some of those solved cases. 

Simulated quantum annealing also slightly outperformed the quantum approach in the reverse-annealing comparison.

The research therefore should not be presented as evidence that quantum computing is ready to operate commercial transportation fleets.

Its value is more specific.

It demonstrates a way to combine historical demand data, real-time fleet information, mathematical optimization, and quantum annealing in a transportation-dispatch model.

For businesses, the next question is scalability.

Future research will need to establish whether these methods continue to provide useful results with larger fleets, real operational data, realistic transportation constraints, and competitive classical alternatives.

For now, the study provides credible evidence that micro-mobility dispatch is a practical research area for quantum optimization—and that the most meaningful progress will come from measuring quantum methods against the real operational requirements of transportation systems rather than from quantum technology alone.