Multi-robot scheduling is where the three fields should meet, and mostly don’t
Task allocation and routing map cleanly onto quantum optimisation. Classical heuristics keep winning at the sizes that matter, and it is worth being precise about why.
Assigning tasks to a fleet, routing vehicles through a warehouse, sequencing a manipulator’s picks — these are combinatorial optimisation problems, and they map onto the quadratic unconstrained binary form that annealers and variational quantum optimisers consume. The mapping is genuine, which is why this is the most-studied overlap of the three fields.
The results are more sobering than the mapping suggests. At the problem sizes that appear in real facilities, well-tuned classical heuristics — local search, large-neighbourhood search, mature mixed-integer solvers — remain ahead on solution quality per unit of wall-clock time. Comparisons that appear to favour quantum approaches often do so against a weak classical baseline, and the field has become appropriately sceptical of that pattern.
There are structural reasons beyond hardware maturity. Encoding constraints into a penalty-based objective inflates the problem and blurs the constraint boundary the solver most needs to respect. Limited qubit connectivity forces minor-embedding that consumes physical qubits at an unfavourable rate. And real scheduling is dynamic — an order arrives, a robot faults — where the relevant capability is fast repair of an existing solution rather than a cold solve.
None of this makes the research direction wrong. It makes the near-term claim wrong. The reasonable position is that this overlap is where a genuine advantage would first become visible if it arrives, and that it has not arrived, and that the way to tell will be a comparison against a classical baseline someone actually tried to make strong.