• Manufacturing

BMW Group: Quantum Optimization for Test Vehicle Production

BMW Group

By

BMW Group's Quantum Computing Challenge posed real production planning problems to the quantum community, including optimizing pre-production test vehicle configuration. Researchers tackled it with D-Wave's hybrid constrained quadratic model solver.

Key Outcome
Independent researchers solved BMW's test vehicle use case with D-Wave's hybrid CQM solver and found its performance comparable to classical solvers such as CBC and Gurobi. No quantum advantage was demonstrated; the value was an honest, reproducible benchmark on a real industrial problem.

The Problem

Before a new model reaches the production line, BMW Group builds a fleet of pre-production test vehicles. Each required test demands certain vehicle features, and each test vehicle can only carry feature combinations that are actually buildable. Test vehicles are expensive, so the optimization goal is to perform the maximum number of required tests on the minimum number of vehicles, while respecting buildability and scheduling constraints.

This is a hard combinatorial problem of exactly the kind quantum optimization researchers want to test themselves against. Rather than keeping it internal, BMW put it in front of the whole quantum community.

The BMW Quantum Computing Challenge

In 2021, BMW Group launched the Quantum Computing Challenge in collaboration with Amazon Web Services, crowd-sourcing quantum approaches to four real use cases: pre-production vehicle configuration (the test vehicle problem), material deformation in production, sensor placement, and machine learning for automated quality assessment.

More than 70 teams worldwide entered, from startups to enterprise research groups, with access to quantum hardware and simulators through Amazon Braket. For the vehicle configuration use case, the winning team combined researchers from 1QBit, NTT Research, and NTT DATA, who proposed both near-term hybrid quantum-classical strategies and longer-term fault-tolerant approaches.

Putting an Annealing-Style Solver to the Test

The test vehicle use case also drew independent academic work. A 2022 paper by Glos, Kundu, and Salehi formulated BMW’s published problem for D-Wave’s hybrid solvers. Instead of forcing everything into a pure QUBO, they used D-Wave’s constrained quadratic model (CQM) format, which lets constraints be stated directly, and solved it with the Leap hybrid CQM solver, benchmarking against the classical solvers CBC and Gurobi.

To see what this kind of formulation looks like in practice, here is a deliberately simplified illustration of assigning jobs to slots with D-Wave’s Ocean tools. This is teaching code, not BMW’s system, and the real test vehicle problem is far larger and more heavily constrained.

# Simplified illustration: assign 4 jobs to 4 slots as a QUBO
import dimod

n_jobs, n_slots, penalty = 4, 4, 10.0
Q = {}

# Each job goes in exactly one slot
for i in range(n_jobs):
    for t in range(n_slots):
        Q[(f"x_{i}_{t}", f"x_{i}_{t}")] = Q.get((f"x_{i}_{t}", f"x_{i}_{t}"), 0) - penalty
        for t2 in range(t + 1, n_slots):
            Q[(f"x_{i}_{t}", f"x_{i}_{t2}")] = 2 * penalty

# Each slot holds at most one job
for t in range(n_slots):
    for i in range(n_jobs):
        for i2 in range(i + 1, n_jobs):
            Q[(f"x_{i}_{t}", f"x_{i2}_{t}")] = 2 * penalty

bqm = dimod.BinaryQuadraticModel.from_qubo(Q)

In a hybrid solver, classical preprocessing decomposes the problem, the quantum annealer explores the energy landscape of suitable subproblems, and classical postprocessing assembles a final solution. That architecture is what makes industrial-scale inputs tractable at all on today’s hardware.

The Honest Result

The researchers’ conclusion was measured: the hybrid CQM solver’s performance was comparable to classical solvers in optimizing the number of test vehicles. Not faster, not better, comparable. No hours-to-minutes speedup exists in the public record, and no quantum advantage was claimed.

That might sound like an anticlimax, but it is exactly what useful near-term quantum research looks like. A real industrial problem was formulated rigorously for quantum hardware, run on the best available hybrid solver, and benchmarked honestly against strong classical baselines. The result tells the field precisely where it stands.

What This Means for Manufacturing

  • QUBO and CQM reformulation is non-trivial. Mapping business constraints into solver-ready form is real engineering work, and BMW’s openly specified challenge problems made that work reproducible.
  • Hybrid is the only viable mode today. Pure annealing cannot hold industrial problem sizes; classical decomposition does the heavy lifting.
  • Benchmarks beat hype. Comparable-to-classical on a genuine BMW problem is more informative than any unverifiable speedup claim.

Manufacturing remains a strong candidate domain for quantum annealing because scheduling, configuration, and logistics problems have natural QUBO formulations. The BMW challenge established the baseline future hardware will have to beat.

Sources