NETWORK FLOWS / REBALANCING
A recorded synthetic study · 24 vehicles · 4 zones · 120-minute observation.
Where should the
empty cars go?
Move before a demand wave, or respond after riders go unserved. What does better coverage cost?
Synthetic scenario · not operational data
Reconstructing the two recorded arms and checking every event, population and ledger row…
THE WHOLE-RUN TRADE-OFF
Coverage has a travel cost.
These totals describe the entire two-hour run. The replay below shows individual moments.
Same workload. Different zone outcomes.
Pickup waits in seconds, served requests only. Compare unserved riders alongside the percentiles. Percentiles alone do not identify an individual rider’s change.
Synthetic scenario · not operational data
01 / THE MAP
Four zones. One fleet.
Uneven demand.
A vehicle can be available and still be in the wrong zone. Compare its location with the next wave of requests.
Your run. Your moves.
Accepted through 0 minutes. Editing a move does not change results.
An empty batch advances without repositioning. You may deliberately leave a zone with no ready vehicles.
Counts come from the checked ledger. The clock shows ten-minute decision snapshots, not continuous streets or vehicle trajectories. Interzone vehicles count in transit, never in two zones.
● Ready / idle◆ COD: executing pickup, trip or empty move□ VOR: unavailable for dispatch; off the drive network, not off the pavement
Synthetic scenario · not operational data
Follow one vehicle.
Read this vehicle's complete event trace
Read waits beside riders not served.
Waits describe pickups within the observation and deadline. Unserved riders are excluded from those percentiles; their counts remain visible.
Synthetic scenario · not operational data
Inspect the full zone ledger and record identity
Boundary rows are before the new decision. Counts include expirations exactly at the right boundary once. Playback uses the separately checked post-decision states. SHA-256 establishes consistency, not independent authenticity.
WHY IT MATTERS / EXTERNAL RESEARCH
Spatial availability
is part of fleet readiness.
Jacobsen, Kulcsár and Lindman's joint dispatch, rebalancing and charging preprint reports up to 36% lower median passenger wait and 27% lower rebalancing distance against its stated baselines. Those are external research results; this small lesson does not reproduce that optimizer, dataset or effect size.
What is fixed, and what does this model leave out?
Twenty-four identical vehicles start A8/B8/C2/D6 on a two-kilometer square. A frozen tape contains 163 requests over twelve ten-minute intervals. Local pickup takes 60 seconds and 200 empty meters. Adjacent-zone travel takes four minutes; diagonal travel takes eight. Pickup deadlines are six minutes after request arrival.
Reactive moves target expirations in the preceding interval. Forecast moves target perfect next-interval origin counts. Both retain at least one ready vehicle in a donating zone, but differing targets also change their reserves. This is a comparison of two transparent heuristics, not only a timing switch, and neither is an optimizer or a trained forecast.
Dispatch is local: no automatic cross-zone pickup. Passenger trips move vehicles between zones. There is no pooling, street traffic, energy model, cost calibration, safety model or VOR event in this fixture. A forecast of nine requests does not imply nine simultaneous cars are necessary.
The independent verifier checks mandatory event coverage, legal allocations, exact request accounting, distance, row coverage and common-input identity. A failed candidate is rejected; its metrics do not replace accepted results.
Authenticity: NOT_AUTHENTICATED. Scope: SYNTHETIC_TEACHING_ONLY. Operational deployment permission: NONE. These results do not establish real-world safety or operator performance.
A SHARED VOCABULARY
Name the state.
Define the denominator.
- Ready / idle
- Available for dispatch, at a zone and not executing a move.
- COD: Car on Drive
- Executing a pickup, passenger trip or empty repositioning move.
- VOR: Vehicle Off-Road
- Unavailable for dispatch: off the drive network, not off the pavement. Charging, updating and maintenance are examples. R1 schedules none of these activities.
- Deadhead ratio
- Empty distance ÷ total distance in matching units. Empty kilometers alone are not this ratio.
- Service rate
- Pickups within the observation and deadline ÷ admitted requests. Completed trips are reported separately.
- Readiness
- Operable-and-ready time ÷ scheduled time, with activity scope defined. This lesson does not calculate that score.
Ready / idle, COD and VOR are FleetLab teaching labels, not a claim of universal AV-industry terminology.
One word, several ratios
Utilization can mean loaded distance ÷ total fleet distance, used link rate ÷ link capacity, or charger power draw ÷ power ceiling. Throughput needs units: bits/s, trips/h or vehicles/h. Service rate here is a served fraction; queueing theory also uses jobs completed per unit time. Name the resource behind capacity. Rebalancing here means empty interzone repositioning; local pickup approach is counted separately.