Quantum optimization meets public transit: lessons from the Berlin Quantum Hackathon

On March 5th, 2026, the grand finale of the Berlin Quantum Hackathon took place at the Change Hub Berlin, bringing together quantum computing teams from across Europe to tackle real-world challenges using state-of-the-art quantum hardware and algorithms. The event served as both a technical competition and a showcase of how quantum approaches are beginning to address genuinely complex industrial problems. Prizes took the form of quantum compute credits, and the awards were presented by Staatssekretär Severin Fischer, reflecting the institutional weight the Berlin quantum ecosystem is beginning to carry.

Our team, Beerantum, secured overall 3rd place and 2nd place in the quantum optimization challenge, with a project targeting one of urban mobility's most persistent bottlenecks: bus crew scheduling at scale.

The problem: scheduling as a combinatorial challenge

BVG, Berlin's public transit operator, manages over 16,000 employees and runs routes that the city depends on every single day. Behind that reliability lies an extraordinarily complex scheduling problem. With 150 drivers, multiple bus lines, a four-week horizon, and tens of thousands of shift segments, the number of feasible assignments grows astronomically. Classical approaches handle the hard constraints well enough, but they tend to flatten the human layer entirely: driver preferences, behavioral patterns, and individual availability are typically left out of the optimization loop.

Ignoring that layer has real consequences. BVG's own internal projections point to over 4,300 employees leaving by 2033 due to retirement alone, with voluntary attrition compounding the challenge. Preference-blind scheduling accelerates that trend.

The quantum approach

We framed the problem as a Quadratic Unconstrained Binary Optimization (QUBO), with binary variables encoding driver-to-segment assignments, hard operational constraints encoded as penalties, and soft driver preferences encoded as rewards. The QUBO was solved using Kipu Quantum's Bias-Field DCQO algorithm, running on the Kipu Quantum Hub.

The pipeline extended well beyond the quantum core. A pre-processing stage parsed rotation segments, built conflict graphs, and used DBSCAN clustering over 16 behavioral features to identify driver archetypes, compressing the preference space and reducing API calls by 80%. An Uncertainty Adapter, combining an Isolation Forest anomaly detector with a Gaussian Process demand predictor, determined dynamically whether a given day warranted a fresh quantum re-optimization or could be served by a pre-computed archetype. Post-processing handled feasibility repair and preference scoring, ensuring hard constraints were always satisfied at output.

The result was a classical-quantum-classical pipeline: classical intelligence at the boundaries, quantum search at the core.

Broader implications

One of the most valuable aspects of working on a problem like this is recognizing how far the architecture generalizes. The same QUBO structure maps naturally to hospital shift scheduling, airline crew planning, last-mile logistics, and energy grid dispatch. The uncertainty pipeline requires only a swap of the feature source to transfer across domains.

On the economic side, a conservative 2% scheduling efficiency gain at BVG's scale translates to roughly €18 million per year. Preference-matched scheduling, reducing even a fraction of voluntary attrition, adds a further €2 to 4 million annually. The project demonstrated a credible path from TRL 4 to a production pilot at TRL 6 within 24 months, aligned to Kipu's hardware roadmap.

Looking ahead

The Berlin Quantum Hackathon confirmed that quantum optimization is moving from theoretical promise to practical pipeline. The challenges are real, the datasets are real, and the hardware, while still maturing, is already capable of anchoring meaningful hybrid workflows.

The experience reinforced something important: quantum systems that work in practice are systems designed with both operational rigor and human complexity in mind. Getting that balance right is where the most interesting engineering problems live.

A bridge to ongoing research

The optimization challenges explored at the hackathon connect naturally to work underway closer to home. Through the Q-Mind project, GMV is investigating quantum algorithms for complex planning tasks, including route optimization, robot scheduling, and satellite constellation coordination, with the broader goal of making real-time solutions viable in domains where classical approaches hit their limits. The crew scheduling pipeline we built in Berlin, with its QUBO formulation, uncertainty handling, and hybrid classical-quantum structure, aligns closely within that research direction. It is a small but concrete illustration of where quantum optimization is heading: from proof-of-concept competitions to integration into the planning systems that critical infrastructure actually depends on.