{"id":"1ce2c33d-d45c-4573-8945-d3a0fdce103c","arxiv_id":"1908.07902","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":5,"one_line_summary":"A rollout-based controller for airport electric baggage trucks cuts simulated daily operating cost by roughly 10% versus greedy charging by shifting charges to solar and off-peak grid power.","lead":"This paper models airport baggage-hauling electric vehicles as a dynamic system and uses a rollout algorithm to schedule charging and luggage runs. A Bristol Airport simulation study reports about 10% lower operating cost than a greedy charging heuristic, mainly by shifting charging to cheap solar and off-peak grid power.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 4 Step 2's full enumeration of next states for 25 ASEVs is not shown tractable; without pruning or decomposition the 'curse of dimensionality' claim is unsupported.","rationale":"The reader's weakest assumption—that Section 4 Step 2's enumeration of all feasible next states is intractable for 25 ASEVs—is exactly the load-bearing concern. The paper's headline contribution is not the 10% cost saving (which is plausible as a rollout improvement over a greedy base policy) but the claim that the customized rollout 'effectively overcomes the curse of dimensionality.' That claim is unsupported because the described algorithm, taken literally, requires minimizing over a combinatorial set of next states with no stated pruning or decomposition. The paper provides no complexity analysis, no bounds on |N(S_t)|, and no code or runtime measurements, so the reader cannot verify that Step 2 is executable. This is an internal gap in the algorithm description, not a disagreement with external consensus. The sign error in Eq. (1) noted by the reader is a separate correctness issue but is secondary to the tractability problem. Because the primary claim of scalability fails under scrutiny, the REJECT verdict remains appropriate; no adjustment is needed.","tokens_in":12504,"tokens_out":3105,"duration_ms":33604,"concrete_test":"Implement Section 4 Step 2 exactly as specified for the 25-ASEV summer scenario and log the cardinality of N(S_t) at each of the 288 stages. If any stage has |N(S_t)| larger than, say, 10^6, or cannot be enumerated within a 5-minute stage budget, the algorithm as written is intractable. Alternatively, run the same rollout with 5, 10, 15, 20, and 25 vehicles and plot per-stage runtime; exponential growth in vehicle count would confirm that the 'curse of dimensionality' has not been overcome.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the rollout 'effectively overcomes the curse of dimensionality' depends entirely on Section 4 Step 2 being executable. As written, Eq. (19)-(20) minimize over N(S_t), 'the set of all possible states at Stage t+1' generated by enumerating all feasible control decisions u_it for every ASEV. With 25 vehicles and three actions per vehicle, the unfiltered product set has 3^25 ≈ 8.5×10^11 elements; the feasibility constraints (9)-(10) and work-in-progress transitions (16)-(17) prune only those vehicles already working or at SoC bounds. The paper gives no bound, pruning rule, decomposition, or approximation of N(S_t), and no complexity analysis. Each candidate also requires simulating one of two heuristics over the remaining horizon to obtain J(S), which multiplies the per-stage cost. The simulation results may be a valid demonstration that rollout improves on greedy charging—this is expected because rollout uses the greedy heuristic as a base policy—but they do not establish that the enumerated search can be performed for a 25-ASEV fleet. If Step 2 cannot be executed, the near-optimal control strategy and the case-study numbers are not reproducible from the described algorithm; the scalability contribution therefore rests on an unverified enumeration assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a dynamics model for airport service electric vehicles (ASEVs) and a rollout-based near-optimal control strategy that jointly decides battery charging and luggage-transport task assignment. The model represents each ASEV by a discrete state (working, charging, idling), a continuous SoC, and a battery-cycle-failure count, with stochastic service-time workload modeled by a discretized truncated normal distribution. The control problem is formulated as a finite-horizon cost minimization over energy purchase, battery degradation, and a terminal SoC penalty. A customized rollout algorithm is built on two base heuristics (renewable-matching and greedy charging), enumerating all feasible next states at each stage and selecting the state with the minimum approximate cost-to-go. Case studies at Bristol Airport with 25 ASEVs on typical summer and winter days report a total daily cost reduction of approximately 10% relative to the greedy-charging heuristic, and a scenario with flight cancellation is used to show adaptability to schedule changes.","tokens_in":12680,"tokens_out":5194,"duration_ms":51219,"significance":"If the central claims hold, the paper would make a useful contribution by addressing a genuinely new application area — centrally controlled airport ground fleets — and by adapting rollout, a classic approximate dynamic programming technique, to a hybrid stochastic system. The ASEV dynamics model with discrete working/charging/idle modes, continuous SoC, and stochastic workload is a reasonable and clearly described formulation. The cost model, including renewable-matching incentives and battery degradation, is plausible and aligned with current interest in grid-friendly EV charging. However, the paper's main quantitative claims — that rollout \"effectively overcomes the curse of dimensionality\" and yields a \"near-optimal\" strategy — are not substantiated by the evidence actually presented. The algorithm's tractability for a 25-vehicle fleet is not demonstrated, the treatment of stochasticity is underspecified, and the absence of an optimality benchmark means the reported cost savings over the base heuristic are expected consequences of rollout's construction rather than evidence of near-optimality.","major_comments":[{"comment":"The enumeration of the set N(S_t) of all feasible next states is not shown to be tractable for the 25-ASEV fleet used in the case study. With three control decisions per vehicle, the unfiltered product set has 3^25 ≈ 8.5×10^11 elements; constraints (9)-(12) and (16)-(17) only exclude vehicles that are already committed to work or at SoC bounds, leaving a potentially enormous set of free vehicles. The paper provides no bound, pruning rule, decomposition, or complexity analysis for N(S_t). Without a tractable enumeration, the central claim that the rollout \"effectively overcomes the curse of dimensionality\" is unsupported, and the algorithm as described may not be executable for the reported case studies.","section":"Section 4, Step 2 (Eqs. (19)-(20))"},{"comment":"The stochastic workload is not integrated into the rollout algorithm in a well-defined way. Transition (11) uses the random variable w_jt, but the rollout description does not state whether J(S) is evaluated using expected values, Monte Carlo samples, scenario trees, or a single realization. The approximate cost-to-go of a candidate next state depends on the future workload distribution, and without an explicit stochastic rollout formulation the numerical results in Section 5 are not reproducible and the algorithm's behavior under uncertainty is not rigorously specified.","section":"Section 4, Steps 1-2 and Section 3.1"},{"comment":"The simulation study is missing several key input parameters and is not reproducible from the manuscript: the truncated-normal workload parameters (μ, σ, a, b) are never specified; the two-tier grid tariff shown qualitatively in Fig. 6 is not given numerically; the battery-degradation cost coefficient is only cited to [23] without a value or formula; and the PV generation profiles are referenced to external websites but not provided in a usable form. In addition, each scenario is a single run of a stochastic process, so the reported 10% cost reduction lacks error bars or a statistical comparison across multiple workload realizations.","section":"Section 5, Scenarios 1-3"},{"comment":"The pseudo-code for the two base heuristic strategies is referenced as Fig. 4 and Fig. 5, but the figures are not present in the manuscript text provided for review. Since the rollout algorithm's construction depends on the exact behavior of these heuristics, the algorithm cannot be implemented or verified from the description alone. The heuristics must be specified in the text, either as pseudo-code or as a precise step-by-step procedure.","section":"Section 4, Figs. 4 and 5"},{"comment":"The claim that the rollout algorithm yields a \"near-optimal\" control strategy is not supported by any comparison with an optimal solution or a lower bound. The approximately 10% cost improvement over the greedy-charging heuristic is a built-in property of rollout: rollout with a base heuristic is designed to improve upon that heuristic, so observing an improvement does not demonstrate closeness to optimality. To justify the near-optimal label, the authors should compare against an exact dynamic-programming solution on a small fleet size, or provide a formal bound on the rollout cost-to-go approximation.","section":"Section 5 and Section 6"}],"minor_comments":[{"comment":"The notation in the truncated normal density is inconsistent: the text states that a and b are the upper and lower bounds, respectively, but the conditions in (1) use \"b < w < a\" and the later integration range [b,a] implies that b is the lower bound and a the upper bound. Please correct the definitions and the conditions in (1).","section":"Eq. (1)"},{"comment":"The sentence \"S_t, N(S_t), and J(S) are defined in (19)\" is slightly inaccurate; J(S) is defined in the text accompanying (19), not in the equation itself. Please rephrase to avoid ambiguity.","section":"Section 4, Eq. (20)"},{"comment":"The statement that the rollout algorithm is \"proven to be adaptive\" is too strong based on a single flight-cancellation scenario; \"demonstrated\" or \"illustrated\" would be more appropriate.","section":"Section 5, Scenario 3"},{"comment":"The tables show control decisions for one example ASEV, but the formatting is difficult to read. Please present the tables with clear column headers (e.g., Start time, End time, Activity) and consider aligning rows for charging and working periods.","section":"Section 5, Tables I-III"}],"recommendation":"reject","confidential_remarks":"The manuscript addresses an interesting and timely application, and the ASEV dynamics model is a reasonable starting point. However, the central scalability claim rests on an unverified enumeration of an astronomically large action space, and the stochastic rollout is not fully specified. The missing pseudo-code and input parameters make the case-study results unreproducible. These are load-bearing issues rather than local presentation problems; a revision would require substantial new algorithmic and experimental content, effectively rewriting the core contribution. I recommend rejection, though I would encourage the authors to resubmit a revised version that addresses tractability, provides a proper optimality benchmark, and fully details the stochastic rollout and all input data."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know this paper's genuine contribution is the ASEV dynamics model, not the rollout algorithm. The model captures discrete work/charge/idle states, SoC dynamics, work-in-progress lock-in, and stochastic service times in a way I haven't seen for airport baggage tractors. The cost function with renewable matching and battery degradation is sensible, and the cancellation case study is a nice touch.\n\nThat said, the headline “overcomes the curse of dimensionality” does not hold up as written. Section 4 Step 2 asks the rollout to enumerate all feasible next states by considering every feasible control decision for every vehicle. With 25 vehicles, even after pruning working vehicles and SoC-bound states, the product set is astronomically large. The waiting-flight constraint (12) forces assignments and helps, but the paper gives no bound, pruning rule, or complexity analysis. The claim is therefore unsupported. The 10% saving over greedy charging is plausible and expected since rollout improves the base policy, but without a benchmark against the true optimum or at least a lower bound, “near-optimal” is self-referential.\n\nThe simulations are also thin: two deterministic days, no error bars, and many parameters (workload moments, tariff values, PV profiles) are only referenced to external sources. Eq. (1) appears to swap the truncation bounds a and b, a typo that should be fixed. These are not fatal, but they make the results hard to reproduce.\n\nI think the reader's take is about right. The model and problem formulation are worth publishing after revision, and the enumeration tractability must be addressed — either with a practical bounding argument, an approximation, or a scaling study. This is for researchers working on EV fleet scheduling and airport ground operations; others can skip. A serious referee should engage with this, but it needs major revision.","headline":"The new ASEV dynamics model is the real contribution; the rollout application is standard and the scalability claim is unsubstantiated, but the paper deserves a major revision.","tokens_in":13269,"tokens_out":4060,"would_cite":false,"duration_ms":40013,"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 customized rollout controller for airport electric baggage vehicles schedules charging to match renewable generation and cuts daily operating cost by about 10% versus a greedy charging heuristic.","keywords":["airport service electric vehicle","energy management","rollout algorithm","near-optimal control","renewable generation matching","battery charging","state of charge","stochastic workload"],"falsifier":"Run the rollout on a small fleet (say four vehicles) where exact dynamic programming is feasible and count the feasible next states at a busy midday stage for 25 vehicles; if the rollout cost exceeds the exact optimum by a wide margin, or if the enumeration cannot finish within a 5-minute stage, the near-optimality or tractability claim fails.","tokens_in":12203,"feed_emoji":"⚡","tokens_out":6892,"duration_ms":65544,"temperature":0.7,"pith_summary":"The paper tries to establish that the hard problem of scheduling a fleet of airport service electric vehicles (ASEVs) — when to charge each battery and which vehicle should serve which flight — admits a near-optimal solution that is actually computable. It proposes a new ASEV dynamics model that mixes discrete work/charge/idle states, a continuous battery state of charge, and stochastic service times, and then applies a customized rollout algorithm that uses two simple charging heuristics as base policies. The central empirical claim is that this rollout policy cuts total daily operating cost by about 10% relative to a greedy \"charge whenever not full\" heuristic on a 25-vehicle Bristol Airport case study for both summer and winter days. The point of caring: electrified airport ground fleets are carbon-intensive per vehicle-hour, and aligning their charging with cheap solar and off-peak grid tariffs reduces both cost and emissions. If true, the method makes matching airport EV load to renewable generation a practical, online decision rule rather than an intractable dynamic program.","feed_headline":"Rollout charging cuts airport EV fleet costs by 10%","feed_subtitle":"A near-optimal controller times baggage-truck charging to match solar PV and off-peak grid tariffs.","key_machinery":"The load-bearing object is the customized rollout algorithm combined with the ASEV dynamics model. The model represents each vehicle by a discrete state $q$ (negative means working with $q$ stages remaining, 0 idling, 1 charging), a continuous state of charge, and a cycle-to-failure count; service times are discrete random variables from a truncated normal distribution. At each stage the rollout algorithm enumerates the set $N(S_t)$ of feasible next fleet states and uses an approximate cost-to-go $\\tilde{J}$, computed by simulating the better of two heuristics to the end of the day, to select the next state that minimizes $\\tilde{J}$. The work this does is to replace the exact dynamic-programming cost-to-go with a heuristic approximation, making a one-step lookahead decision that captures future charging opportunities while remaining online.","core_discovery":"On its own terms, the paper claims that the ASEV energy-management problem can be solved near-optimally by a custom rollout algorithm. Starting from a current fleet state, the algorithm enumerates the feasible next states reachable by choosing charge, idle, or work for each vehicle, estimates the remaining cost-to-go from each candidate by running the better of two heuristics (one that charges only when renewable energy is available, one that charges as soon as possible), and moves to the candidate with the lowest estimated total cost. Repeating this at each 5-minute stage yields an online control strategy that schedules both luggage transport and battery charging. In the Bristol Airport case studies the strategy reduces total daily cost by about 10% compared with greedy charging, achieves this by buying much less grid energy during peak tariff periods, and adapts to flight cancellations on short notice.","pith_inferences":["The same rollout template — discrete task assignment, SoC dynamics, and two base heuristics — should transfer to other centrally dispatched electric fleets such as port equipment, warehouse robots, and shuttle buses whose tasks are stochastic and whose charging is price-responsive.","The paper does not bound the optimality gap; comparing rollout against an exact solution for a small fleet would show how much of the 10% saving comes from the rollout itself rather than from the underlying heuristics.","A safety-SoC constraint that prevents the renewable-matching heuristic from becoming infeasible might preserve most of the PV benefit while avoiding the failure seen in the summer scenario.","Because the saving comes partly from deferring charge to off-peak hours, the method's value depends on tariff spread and PV availability; a sensitivity study across tariff ratios would show where rollout ceases to beat greedy charging."],"forward_implications":["An airport operating under similar two-tier tariffs and PV profiles can expect roughly a 10% daily cost reduction from rollout control instead of greedy charging.","The charging load shifts toward cheap solar and off-peak grid windows, making the fleet's load curve friendlier to local renewable consumption and to the grid.","The receding-horizon rollout re-plans after flight cancellations, so schedules need not be frozen a day ahead.","Battery degradation cost rises because the controller allows deeper discharges, even though total cost falls; the net benefit depends on the relative weight of battery replacement cost."],"supporting_citations":[{"why":"Supplies the rollout and approximate cost-to-go machinery used to select the next fleet state.","marker":"[20]"},{"why":"Defines the truncated normal distribution used to model stochastic flight service times.","marker":"[22]"},{"why":"Provides the coefficient for the linear battery degradation cost function used in the objective.","marker":"[23]"},{"why":"Supplies the Bristol Airport flight schedule used in all three case-study scenarios.","marker":"[24]"},{"why":"Supplies the summer PV generation profile for the first and third scenarios.","marker":"[25]"},{"why":"Supplies the winter PV generation profile for the second scenario.","marker":"[26]"},{"why":"Sets the constant fast-charging power of 22 kW used in the battery dynamics.","marker":"[27]"},{"why":"Sets the 90% battery cycle efficiency used in the charging dynamics.","marker":"[28]"},{"why":"Sets the 50 kWh battery capacity underlying the state-of-charge constraints.","marker":"[29]"}],"fun_headline_variants":["Rollout control trims airport EV fleet costs by 10%","Airport EVs: rollout charging beats greedy by 10%","Near-optimal rollout for airport EV charging cuts costs","Airport service EVs matched to renewables save 10%","Rollout algorithm adapts airport EV charging to solar and tariffs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The algorithm's tractability depends on being able to enumerate the full set of feasible next fleet states at each 5-minute stage; the paper does not prune or decompose the action space, and with 25 vehicles and three actions per vehicle this candidate set can become astronomically large.","fun_headline_variants_meta":{"raw":{"variants":["Rollout control trims airport EV fleet costs by 10%","Airport EVs: rollout charging beats greedy by 10%","Near-optimal rollout for airport EV charging cuts costs","Airport service EVs matched to renewables save 10%","Rollout algorithm adapts airport EV charging to solar and tariffs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000438,"raw_usage":{"total_tokens":2216,"prompt_tokens":930,"completion_tokens":1286,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":1202}},"tokens_in":546,"tokens_out":1286,"duration_ms":9023,"temperature":1.0,"reasoning_tokens":1202,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:54:18.996991+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the rollout on a small fleet (say four vehicles) where exact dynamic programming is feasible and count the feasible next states at a busy midday stage for 25 vehicles; if the rollout cost exceeds the exact optimum by a wide margin, or if the enumeration cannot finish within a 5-minute stage, the near-optimality or tractability claim fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the rollout and approximate cost-to-go machinery used to select the next fleet state."},{"cited_title":"The Truncated Normal Distribution,","cited_arxiv_id":null,"evidence_quote":"Defines the truncated normal distribution used to model stochastic flight service times."},{"cited_title":"A linear programming approach for battery degradation analysis and optimization in offgrid power systems with solar energy integration,","cited_arxiv_id":null,"evidence_quote":"Provides the coefficient for the linear battery degradation cost function used in the objective."},{"cited_title":"Bristol Airport Arrivals and Departures","cited_arxiv_id":null,"evidence_quote":"Supplies the Bristol Airport flight schedule used in all three case-study scenarios."},{"cited_title":"Weather & Local Environment","cited_arxiv_id":null,"evidence_quote":"Supplies the summer PV generation profile for the first and third scenarios."},{"cited_title":"Live monitoring of the UK electricity National Grid","cited_arxiv_id":null,"evidence_quote":"Supplies the winter PV generation profile for the second scenario."},{"cited_title":"Understanding electric car charging","cited_arxiv_id":null,"evidence_quote":"Sets the constant fast-charging power of 22 kW used in the battery dynamics."},{"cited_title":"On Generation Schedule Tracking of Wind Farms With Battery Energy Storage Systems,","cited_arxiv_id":null,"evidence_quote":"Sets the 90% battery cycle efficiency used in the charging dynamics."},{"cited_title":"Electric car boom in Europe starts this year","cited_arxiv_id":null,"evidence_quote":"Sets the 50 kWh battery capacity underlying the state-of-charge constraints."}],"review_version":1}