{"id":"d98ae92a-f481-483f-8995-6095b3d54f92","arxiv_id":"1908.09828","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"In a calibrated Ann Arbor evening-traffic simulation, a fuel-aware ride-sharing assignment objective reduces fuel consumption per customer by up to 7% compared with personal vehicles while serving over 90% of riders within time limits.","lead":"This paper simulates a ride-sharing fleet controller that explicitly minimizes fuel use in its assignments, and reports up to 7% lower fuel per customer than a private-car baseline. The result matters because it also shows that ignoring fuel in fleet control can increase fuel use through empty vehicle miles.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 7% fuel-saving claim may be inflated because the same fuel model is used both as the optimization objective and as the evaluation metric; an independent emissions model should be used to score the configurations.","rationale":"The paper is a careful simulation study, and the algorithmic contribution is plausible. The strongest claim, however, is an empirical number (7%) that is generated and then measured inside the same modeling framework. The most load-bearing assumption is therefore not the congestion independence (which the authors explicitly acknowledge in Section VI and is reasonable at 4% penetration) but the identity between the optimization objective and the evaluation metric. This is a circularity risk, not an ad hominem or consensus disagreement. If the fuel model were clearly validated against independent data, the concern would dissolve; as it stands, no such validation is reported. The reader's weakest assumption (calibration bias) is related but distinct; the rationale does mention the shared fuel model, so my agreement is partial. I would keep the verdict conditional: the paper should be accepted as a methodological framework only if the authors either release code/data for independent replication or re-evaluate with an independent emissions model. Without that, the specific 7% energy-saving claim cannot be taken at face value.","tokens_in":14786,"tokens_out":4998,"duration_ms":53062,"concrete_test":"Recompute fuel consumption for all configurations and the baseline using an independent, validated emission model (e.g., MOVES or SUMO's HBEFA emissions) applied to the exact logged vehicle trajectories from the existing SUMO runs. If the relative reduction of configuration 8 versus configuration 9 (baseline) drops below the calibration uncertainty band or changes sign, the headline 7% claim is an artifact of using the optimized fuel model for evaluation. As a secondary check, perturb the fuel model coefficients by ±1 standard deviation and verify that the ordering of configurations and the 7% figure are preserved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline result (Section V.C) is that fuel-oriented assignment cuts fuel consumption per customer by 1.4–7.0% versus a personal-vehicle baseline. The metric used to measure this is not an independent ground truth: the data-driven fuel model from [3] is used both as the trip-assignment and eco-routing cost (Sections II and IV) and to compute the reported fleet fuel consumption in Figs. 14 and 21–23. No separately validated emissions model (e.g., MOVES or SUMO's HBEFA module) is applied to the simulated trajectories. Optimizing exactly the same function that later scores the system lets the planner exploit the model's specific assumptions—e.g., average-speed-only fuel dependence—so part of the measured gain is an artifact of objective alignment rather than a physical fuel saving. The risk is compounded by the reported calibration bias (mean speed error −1%, standard deviation 25%, and visibly less downtown congestion, Section V.A), which means the fuel model's inputs are not representative of the real network. Consequently, the 7% figure is a property of the coupled optimizer+model, not yet a property of the algorithm in the world.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops Eco-MOD, a fleet control algorithm for shared mobility-on-demand that minimizes fleet fuel consumption while satisfying customer wait-time and delay constraints. The algorithm extends the shareability-graph and clique-based assignment framework of Alonso-Mora et al. (2017) by replacing travel-time costs with fuel-consumption costs, using a data-driven fuel model from the authors' prior work, and it adds an active idle-fleet rebalancing step. The system is evaluated in a SUMO microsimulation calibrated with SPMD data and POLARIS-generated demand for Ann Arbor, comparing eight configuration of assignment cost (travel time vs. fleet fuel) and routing strategy (fastest, eco, or hybrid) against a personal-vehicle baseline. The headline finding is that fuel-aware assignment with 1.2-1.3 passengers per vehicle reduces fuel consumption per customer by 1.4-7.0% relative to the baseline while serving over 90% of customers within constraints, whereas travel-time-oriented assignment can increase fuel consumption by 13% due to empty-vehicle mileage.","tokens_in":15063,"tokens_out":2983,"duration_ms":33247,"significance":"If the result holds, the paper makes a useful and practical contribution: it demonstrates that including fuel consumption directly in the trip-assignment and routing objective, rather than relying on travel-time optimization or purely demand-side ride-sharing, can reduce fleet-level fuel use. The simulation framework is carefully structured, with real-world calibration data, a microscopic traffic simulator, and a clear set of controlled configurations. The algorithmic extension of the shareability-graph method to fuel costs is plausible and well described. The main significance is conditional on the validity of the evaluation, because the headline 7% figure is produced inside a simulator with a fuel model that is also the optimization objective. The paper would be substantially strengthened by an independent fuel/emissions evaluation, a sensitivity analysis of the calibration error, and additional scenarios or replications.","major_comments":[{"comment":"The reported fuel savings are measured with the same data-driven fuel consumption model [3] that is used as the optimization objective in the trip assignment (Section II.A) and as the routing cost (Section IV). Because the planner minimizes exactly the function that later scores the configurations, part of the measured gain may be an artifact of objective alignment rather than a physical fuel reduction. For example, the model's average-speed-only dependence may let the optimizer choose routes and assignments that look fuel-efficient under the model but would not transfer to a more detailed emissions model. I request an evaluation of the same simulated trajectories and assignments with an independent fuel consumption or emissions model (e.g., MOVES, SUMO's HBEFA-based model, or measured second-by-second fuel data) to confirm that the 1.4-7.0% savings are not an artifact of the shared model.","section":"Section V.A"},{"comment":"The traffic calibration reports a mean relative speed error of only -1% but a standard deviation of 25%, and the authors state that the simulation shows less congestion in the downtown area than measured values, possibly because pedestrians and public transit are not modeled. Since the fuel model consumes link speeds, a 25% spread in speed error and a systematic downtown bias can shift both the baseline personal-vehicle fuel and the MOD fleet fuel, and the directional effect on the 1.4-7.0% savings is not quantified. Please add a sensitivity analysis that perturbs the calibrated speeds or demand within the reported error distribution and reports the resulting range of the per-customer fuel savings, or re-calibrate the downtown network with additional modes.","section":"Section V.A"},{"comment":"The headline result is based on a single scenario: one city (Ann Arbor), one time window (17:00-19:00), one MOD penetration ratio (4%), one fleet size (1,200), and no reported statistics over multiple random seeds or demand draws. The paper states in Section VI that the analysis assumes the MOD penetration is small enough that its effect on link travel times is negligible, but this assumption is not tested. To establish that the 1.4-7.0% reduction is a property of the algorithm rather than a single calibrated case, please report run-to-run variability (e.g., mean and confidence intervals over random seeds) and, if feasible, results for at least one additional demand scenario or a sensitivity sweep over penetration ratio within the current simulation framework.","section":"Section V.C"}],"minor_comments":[{"comment":"In the '9' row of Table I, 'Shorest Distance/Fastest' contains a typo; it should read 'Shortest Distance/Fastest.'","section":"Table I"},{"comment":"The phrase 'due to the the lower trip average speed' contains a duplicated 'the'; please correct it.","section":"Section V.C"},{"comment":"The abbreviation MOD is introduced as 'Mobility-on-Demand' in the abstract but later used as 'mobility-on-demand (MOD)' in the introduction; please make the usage consistent.","section":"Section II"},{"comment":"The paper claims the algorithm is 'fast enough for real-world implementation' but does not report computation times for the TSP/ILP/rebalancing steps. A brief timing table would support that claim.","section":"Section V.C"}],"recommendation":"major_revision","confidential_remarks":"The paper's central idea is sound but its headline estimate is not yet convincing because the evaluation metric coincides with the optimization objective and the calibration error is substantial. The authors should be asked to re-score their configurations with an independent fuel model and to provide sensitivity analysis. I would also encourage the editor to verify that the 7% claim is not overinterpreted in the abstract, since the paper itself notes the assumptions on fleet penetration and congestion effects. The novelty relative to the authors' earlier work [41] is incremental but acceptable for a journal publication if the validation gap is closed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this paper shows that putting fuel consumption directly in the assignment objective of a ride-sharing MOD controller can flip the fuel balance relative to a personal-vehicle baseline in a realistic Ann Arbor simulation. The reported 7% reduction is plausible but should not be read as a physical guarantee until an independent fuel model scores the trajectories.\n\nWhat's genuinely new: the authors take the Alonso-Mora shareability-graph framework, which is travel-time oriented, and replace the assignment cost with a data-driven fuel cost. They also integrate POLARIS demand generation with SUMO microsimulation and a simple calibration to SPMD data. The simulation study then demonstrates that a time-only assignment can increase fleet fuel by 13% because of empty vehicle miles, while a fuel-aware assignment with ride-sharing brings it down 1.4-7% below a personal-vehicle baseline. That gives operators a concrete design lesson: the control objective matters, and ignoring fuel is not neutral.\n\nThe soft spots are real but not fatal. First, the same fuel model [3] is used both as the optimization cost and as the evaluation metric. The optimizer is pushing on the exact function that later reports the headline number, so part of the gain could be the optimizer exploiting the model's average-speed-only structure rather than a physical reduction. This is not full circularity, but it is a valid reason to discount the certainty. Second, the calibrated simulator has a mean speed error of -1% with a 25% standard deviation and visibly less downtown congestion. The paper acknowledges these limitations, which helps, but they still shift the baseline and the fleet estimates. Third, this is one city, one evening peak, one demand penetration ratio; the 7% is a data point, not a law. No code or data is released.\n\nWho gets value: researchers working on MOD fleet control, eco-routing, and simulation-based evaluation of shared mobility. The paper is clear enough to serve as a reference implementation of a fuel-aware assignment in an established framework.\n\nMy recommendation: yes, send it to peer review. The core formulation and simulation are described in enough detail to be scrutinized, the result is new in a practical sense, and the weaknesses are quantifiable and addressable. A referee should ask for an independent fuel/emissions model (e.g., MOVES or SUMO's HBEFA) and ideally a sensitivity analysis on the demand ratio and calibration parameters.","headline":"A useful, well-described simulation study showing fuel-aware ride-sharing assignment can reduce fleet fuel in a calibrated Ann Arbor model, but the headline 7% is probably optimistic until an independent fuel model is used for evaluation.","tokens_in":15569,"tokens_out":2488,"would_cite":true,"duration_ms":25370,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A fuel-minimizing ride-sharing fleet controller cuts fuel per customer by up to 7% in simulation.","keywords":["mobility-on-demand","ride-sharing","fleet control","fuel consumption optimization","eco-routing","dynamic vehicle routing","empty vehicle mileage","traffic simulation calibration"],"falsifier":"Re-run the same assignment controller with link travel times that respond to the fleet's own trips at the assumed fleet penetration; if the fuel saving per customer relative to personal vehicles disappears or reverses, the central claim fails. A field alternative is to measure empty-vehicle mileage in an operating ride-sharing fleet: the claim predicts fuel-oriented assignment cuts empty miles enough to undo the 13% penalty that time-only assignment produces.","tokens_in":14626,"feed_emoji":"⛽","tokens_out":6863,"duration_ms":68657,"temperature":0.7,"pith_summary":"This paper tries to establish that a shared automated vehicle fleet can save fuel directly by making fleet fuel consumption the objective of trip assignment, rather than treating travel time as the only cost. In a calibrated simulation of evening rush-hour demand in a mid-sized city, the proposed Eco-MOD algorithm assigns 1.2 to 1.3 passengers per vehicle on average and reduces fuel consumed per customer by 1.4% to 7.0% compared with a personal-vehicle baseline, while more than 90% of customers are served within wait-time and delay-time constraints. The paper also shows the failure this fixes: a fleet assigned purely by travel time can increase fuel consumption per customer by up to 13%, because empty rebalancing miles outweigh the benefit of sharing. The result matters because it says ride-sharing's energy benefit is not automatic; it depends on putting fuel cost into the assignment controller.","feed_headline":"Simulated ride-share fleet cuts fuel per rider by up to 7%","feed_subtitle":"Assigning shared trips by fuel cost beats personal-car fuel use while still serving 90% of riders on time","key_machinery":"The load-bearing machinery is a shareability-graph decomposition of the ride-sharing problem. Nodes are customers and vehicles; an edge means a virtual vehicle can serve both endpoints without violating wait and delay limits, and feasible serving patterns are cliques. Each clique is solved exactly as a small pickup-and-delivery traveling salesman problem, with fuel consumption as the transitional cost, and the resulting candidate trips are selected by integer linear programming. This turns a hard dynamic vehicle-routing problem into many small optimizations. Around this core sits a rebalancing mechanism that relocates idle vehicles toward the trip-origin distribution with fuel as the relocation cost, and that mechanism is what accounts for the empty vehicle miles that otherwise inflate fuel use.","core_discovery":"The central claim is that a ride-sharing fleet should be controlled with fuel consumption as the assignment cost, not as an afterthought. When the trip-assignment objective is total fleet fuel and each vehicle carries about 1.2–1.3 assigned customers on average, the simulated fleet uses 1.4% to 7.0% less fuel per customer than the baseline in which every trip is made by a personal vehicle. Under the same calibrated simulation, a travel-time-oriented fleet burns up to 13% more fuel per customer than the baseline because of empty vehicle mileage from rebalancing. The paper further finds that eco-routing adds 5–8% fuel reduction on top of the assignment-level gains, and that all tested configurations serve more than 90% of customers within their travel time constraints.","pith_inferences":["If the same controller were run with congestion feedback, the 7% figure would change; the paper's own scope condition says the fleet is small, so a scaled-up fleet is the natural stress test of the result.","For electric fleets, swapping the fuel model for an energy-and-charging cost would likely preserve the structural result, but empty-mile penalties and rebalancing costs would remain and should be re-measured.","Per-customer fuel metrics that ignore empty rebalancing distance can overstate the benefit of ride-sharing; the paper's 13% penalty demonstrates why empty miles belong in the numerator.","The 1.4–7.0% range is tied to the demand pattern and time constraints used; denser demand or looser delay limits could push occupancy and savings outside this range."],"forward_implications":["Ride-sharing operators that optimize only wait and delay can burn more fuel per customer than personal cars; adding fuel cost to the assignment is the corrective.","Fuel-oriented assignment works by raising occupancy to 1.2–1.3 customers per vehicle, so sharing, not eco-driving alone, is the main fuel lever.","Eco-routing adds a further 5–8% fuel reduction on top of either assignment objective, so routing and assignment should be tuned together.","Fuel-oriented fleets need slightly more vehicles than time-oriented fleets to keep 90% of customers within time constraints, because higher occupancy raises wait and delay.","A weighted objective that balances system fuel against individual time cost can trace a Pareto frontier, with one tested configuration offering good time performance and fuel savings."],"supporting_citations":[{"why":"Supplies the shareability-graph and integer-programming ride-sharing assignment framework that Eco-MOD extends with fuel costs.","marker":"[8]"},{"why":"Supplies the data-driven fuel consumption model used as the trip-level cost and for eco-routing.","marker":"[3]"},{"why":"Generates the travel demand prior used to define customer origins and destinations.","marker":"[35]"},{"why":"Provides the real driving-speed data used to calibrate the traffic simulator.","marker":"[36]"},{"why":"Provides the microscopic traffic simulator used to evaluate fleet control and measure fuel outcomes.","marker":"[37]"},{"why":"Supplies the road-network partitioning and trip-origin distribution model used for idle-fleet rebalancing.","marker":"[41]"},{"why":"Provides a queuing-network model used to estimate the fleet size needed for the simulation.","marker":"[47]"}],"fun_headline_variants":["Fuel-cost dispatch cuts ride-share fuel use 7%","Eco-routing and fuel-aware trip assignment save 7%","Time-first fleets burn more; fuel-first fleets save 7%","Ride-share control that prices fuel trims consumption"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The headline fuel saving comes from a calibrated simulator in which the ride-sharing fleet is assumed to be small enough not to change road congestion; if the fleet's driving alters link travel times, or the calibration misrepresents downtown congestion, the baseline and fleet fuel numbers that generate the 7% saving would shift.","fun_headline_variants_meta":{"raw":{"variants":["Fuel-cost dispatch cuts ride-share fuel use 7%","Eco-routing and fuel-aware trip assignment save 7%","Time-first fleets burn more; fuel-first fleets save 7%","Ride-share control that prices fuel trims consumption"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000137,"raw_usage":{"total_tokens":1074,"prompt_tokens":795,"completion_tokens":279,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":411,"completion_tokens_details":{"reasoning_tokens":205}},"tokens_in":411,"tokens_out":279,"duration_ms":3594,"temperature":1.0,"reasoning_tokens":205,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:25:04.880568+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the same assignment controller with link travel times that respond to the fleet's own trips at the assumed fleet penetration; if the fuel saving per customer relative to personal vehicles disappears or reverses, the central claim fails. A field alternative is to measure empty-vehicle mileage in an operating ride-sharing fleet: the claim predicts fuel-oriented assignment cuts empty miles enough to undo the 13% penalty that time-only assignment produces.","supporting_citations":[{"cited_title":"On-demand high -capacity ride -sharing via dynamic trip -vehicle assignment,","cited_arxiv_id":null,"evidence_quote":"Supplies the shareability-graph and integer-programming ride-sharing assignment framework that Eco-MOD extends with fuel costs."},{"cited_title":"Eco -Routing based on a Data Driven Fuel Consumption Model,","cited_arxiv_id":null,"evidence_quote":"Supplies the data-driven fuel consumption model used as the trip-level cost and for eco-routing."},{"cited_title":"POLARIS: Agent-based modeling framework de velopment and implementation for integrated travel demand and network and operations s imulations,","cited_arxiv_id":null,"evidence_quote":"Generates the travel demand prior used to define customer origins and destinations."},{"cited_title":"Safety pilot model deployment: Test conductor team report,","cited_arxiv_id":null,"evidence_quote":"Provides the real driving-speed data used to calibrate the traffic simulator."},{"cited_title":"Recent Development and Applications of SUMO–Simulation of Urban MObility,","cited_arxiv_id":null,"evidence_quote":"Provides the microscopic traffic simulator used to evaluate fleet control and measure fuel outcomes."},{"cited_title":"Efficient Mobility -on-Demand System with Ride-Sharing,","cited_arxiv_id":null,"evidence_quote":"Supplies the road-network partitioning and trip-origin distribution model used for idle-fleet rebalancing."},{"cited_title":"Robotic load balancing for mobility-on-demand systems The,","cited_arxiv_id":null,"evidence_quote":"Provides a queuing-network model used to estimate the fleet size needed for the simulation."}],"review_version":1}