Replication of Volkswagen / D-Wave 2017 (arXiv:1708.01625): route assignment on T-Drive data — minimize congested roads. One polycontextural IBM submission (3 zones + transjunctions) vs sequential zone jobs.
The original study needed qbsolv because ~1,254 variables did not fit one D-Wave 2X call — classical partitioning with no quantum coupling between partitions. QPC’s thesis: geographic contextures coupled by transjunctions on shared road segments in one gate-model run.
| Workflow | Submissions | Coupling across zones |
|---|---|---|
| qbsolv (D-Wave 2017) | Many | Classical partition loop |
| Sequential 3-zone QAOA | 3 | Merge only after separate jobs |
| Weighted QAOA (single QUBO) | 1 | Full QUBO, no zone structure |
| QPC polycontextural | 1 | Transjunctions on shared segments |
| Method | Congested roads | Notes |
|---|---|---|
| Random assignment | ~120–160 | Fig. 4 boxplot (418 cars) |
| qbsolv + D-Wave 2X | ~40–80 | Hybrid decomposition |
| Original routes | Baseline line in Fig. 4 | T-Drive GPS paths |
Data: Microsoft T-Drive · Paper does not publish exact Q matrix — we use faithful reconstruction.
Instance: 16 cars · 48 route variables · 6×6 synthetic grid with shared arterial · 8192 shots · QPC noise reducer · 5 IBM jobs (full baseline suite)
QPC decoder: 2 congested roads — matches greedy, weighted QAOA, and 3-zone sequential merge (2 each); beats random mean (7.46).
Architecture: same score as qbsolv-style 3-zone hybrid, but QPC uses 1 coupled submission vs 3 separate zone jobs.
| Method | Congested ↓ | IBM jobs | Job ID(s) |
|---|---|---|---|
| Greedy / tabu classical | 2 | 0 | — |
| Random (mean of 50) | 7.46 | 0 | — |
| Weighted QAOA | 2 | 1 | d8gq8mpe8nrc73bgcalg |
| Sequential zone 0 / 1 / 2 | 2 each | 3 | d8gqaom6983c73dq7qh0 · d8gqcre6983c73dq7tr0 · d8gqg55v8cos73f3jar0 |
| QPC polycontextural | 2 | 1 | d8gq8q42upec739k6skg |
Earlier QPC-only run: d8gq6m42upec739k6pig · JSON: results/traffic_compare_ibm_fez_16c_full.json
Heron156 mode pads to 3×52Q when n_vars ≤ 52 for full-chip campaigns; this pilot uses 48Q directly on Fez.
cd site_release_2025_11_15
.venv/bin/python vendor_benchmarks/traffic/test_pipeline.py
.venv/bin/python vendor_benchmarks/traffic/qpc_traffic_compare.py --mode ibm --backend ibm_fez --n-cars 16 --grid 6 --shots 8192