{"id":"4bc8a835-d73c-4fcf-8729-40597607af5c","arxiv_id":"2411.09783","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"A mixed-integer optimization framework combines vehicle routing and DC optimal power flow to dispatch autonomous vehicles as mobile grid batteries.","lead":"This paper proposes an optimization framework that routes idle autonomous vehicles to power grid nodes so their onboard batteries can help balance supply and demand. It frames self-driving electric vehicles as mobile storage units that could lower grid operating costs without building new infrastructure.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Routing MILP (6) does not force the route to reach the destination: Eq. (2) only forbids outgoing edges from e, so a directed cycle through s can be feasible and cheaper than any s–e path.","rationale":"The paper's central claim has two pillars: the exact McCormick reformulation and the optimal routing of vehicles. The McCormick pillar is sound, because z is binary and the inequalities (13a)-(13d) are exact. The routing pillar, however, is not. Every travel cost in (14a) is computed by solving (6), so a flaw there propagates into the dispatch and the reported 26.27% cost saving. The reader's weakest_assumption (Assumption 1: ignoring battery SOC and discharge efficiency) is a legitimate acknowledged simplification and is important, but it is disclosed and deferred to future work. The missing arrival constraint is unacknowledged and mathematical; it breaks the stated optimization problem. A one-line fix is available, so the paper is conditionally salvageable, but the current text cannot be accepted as a correct formulation. I therefore recommend CONDITIONAL rather than UNCHANGED, and I disagree with the reader's choice of weakest assumption.","tokens_in":7445,"tokens_out":11220,"duration_ms":119735,"concrete_test":"Solve problem (6) as written on the four-node instance: s=1, e=4, edges 1->2, 2->3, 3->1 with weight 1 each, and 1->4 with weight 100. Since the cycle 1->2->3->1 is feasible and costs 3, while the path to 4 costs 100, a correct s–t path formulation must return 100; if the solver returns 3 or any solution with no incoming edge to node 4, the missing arrival constraint is confirmed. Then add sum_{i in A_e} x_{i e}=1 and re-run; the optimum should become 100.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The load-bearing flaw is in the routing layer. Problem (6) is supposed to return the shortest route from s to e, and its output c_v feeds the objective of the dispatch MILP (14). But constraints (1)–(5) never require the destination to be reached. Eq. (2) only sets outgoing edges from e to zero, and flow conservation is imposed only for nodes other than s and e. No constraint forces an incoming edge to e, and no constraint forbids incoming edges to s. Consequently a directed cycle through s, e.g., s->a->b->s, satisfies all constraints while never visiting e. With positive branch weights this can even be cheaper than the true s–e path. In a 4-node graph with s=1, e=4, edges 1->2, 2->3, 3->1 each of weight 1, and edge 1->4 of weight 100, the cycle has cost 3 and is feasible, while the shortest s–e path costs 100. Thus the 'optimal route' and travel cost c_v in Eq. (7) are not necessarily routes to the nominated candidate node, so the power-system dispatch built on those costs is not grounded. The missing constraint is sum_{i in A_e} x_{i e}=1 (and typically also sum_{i in A_s} x_{i s}=0). This is an internal inconsistency, not a modeling simplification.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a two-layer framework for dispatching idle autonomous unmanned vehicles to power-system buses so that their onboard batteries can supply energy and reduce operating cost. The transportation layer is a shortest-path MILP (problem (6)), and the power layer is a DC optimal power flow with vehicle-dispatch binaries and a McCormick linearization of the bilinear product p_v z_i^v (problem (14)). Numerical experiments on the IEEE 14-bus system report a 26.27% cost reduction, and additional tests up to a 1354-bus system report solve times under 10 seconds.","tokens_in":7678,"tokens_out":7205,"duration_ms":72945,"significance":"If the result holds, the paper addresses a timely and relevant problem: using autonomous vehicles as mobile batteries in grid operations. Its main mathematical contribution, replacing the bilinear term p_v z_i^v with the McCormick inequalities (13a)-(13d), is correct and exact because z_i^v is binary; this is a genuine strength. The two-layer decomposition is clean, and the scalability table (up to 1354 buses in under 10 seconds) is useful evidence of computational tractability. However, the routing-layer correctness gap and the idealized battery assumption mean the validity of the downstream claims is not yet established. With the routing fix and stronger numerical validation, the framework would be a useful addition to the mobile-energy-storage literature.","major_comments":[{"comment":"The routing formulation is missing an arrival constraint. Eq. (2) only forbids outgoing edges from the destination e; nothing forces an incoming edge to e or forbids incoming edges to s, and flow conservation in Eq. (3) is imposed only for nodes other than s and e. As a result, a directed cycle through s, e.g., s→a→b→s in a four-node graph, satisfies all algebraic constraints while never visiting e, and with positive weights it can be strictly cheaper than the true shortest s–e path. Since the resulting c_v from Eq. (7) feeds the dispatcher objective (14a), every downstream dispatch claim is affected. The fix is to add ∑_{i∈A_e} x_{i e}=1 and ∑_{i∈A_s} x_{i s}=0, and to add subtour-elimination constraints if zero- or negative-weight edges are to be allowed. This is an internal inconsistency in the optimization model, not a modeling simplification.","section":"Section II, Eqs. (1)-(6)"},{"comment":"The model assumes each dispatched vehicle can reliably inject its nominal power p_v for the entire support horizon. Because state of charge and discharge efficiency are ignored, the power balance equality (14c) may be infeasible in practice even when the MILP reports a feasible solution. The paper acknowledges this simplification and defers it to future work, but the headline results (the 26.27% cost reduction and the feasibility of dispatch) currently rest on this assumption. I ask the authors to either incorporate battery-energy constraints (or a conservative energy limit) in the dispatch model, or to reframe the quantitative claims as an idealized upper bound and add a sensitivity study with respect to available vehicle energy.","section":"Section III, Assumption 1 and Eq. (14c)"},{"comment":"The numerical evidence for the economic benefit is based on a single synthetic transportation network with weights randomly drawn between 1 and 12 and cost coefficients 'designed' so that neither term dominates. No comparison is made against a no-vehicle baseline, stationary batteries, or the mobile-storage formulations cited in [5], [6], and no sensitivity analysis is reported. The 26.27% cost reduction should therefore be presented as an illustrative example rather than a validated performance claim. I recommend adding multiple random instances, a small sensitivity study over weights and cost coefficients, and at least one baseline comparison.","section":"Section IV, IEEE 14-bus results and Table I"}],"minor_comments":[{"comment":"The notation R_v = {(i,j) | x*_ij = 1} is correct only if the routing problem is repaired so that the selected edges actually form an s–e path; please add a sentence clarifying that this set is the intended route after the missing constraints are included.","section":"Section II, paragraph after Eq. (7)"},{"comment":"The output list includes e_v, but e_v is not defined until Eq. (15); please move or define the notation before the algorithm or add a pointer to Eq. (15).","section":"Algorithm 1"},{"comment":"Table I reports only the solve time for the dispatch MILP (14); please state explicitly whether the offline routing solves of problem (6) are included in any of the reported times, and if not, what their total computational cost is.","section":"Table I and Section IV"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here is the thing you need to know: the routing formulation in (6) does not require the vehicle to reach its destination. Constraint (2) only forbids outgoing edges from e. There is no constraint that forces an incoming edge to e, and incoming edges to s are not forbidden. As a result, a directed cycle through s (s→a→b→s) is feasible, and with positive edge weights it can be cheaper than any actual s–e path. The \"optimal route\" and the travel cost c_v in (7) may therefore have nothing to do with reaching e, and the dispatch problem (14) uses those costs as coefficients. This is a genuine, load-bearing flaw, not a simplification. The fix is simple: add Σ_{i∈A_e} x_{ie}=1 and typically Σ_{i∈A_s} x_{is}=0, but without it the method does not do what the paper claims.\n\nWhat works: the power-system formulation is standard but correctly assembled. The binary placement variables, the DC power flow, and the McCormick relaxation are all fine; the exactness argument for (13) is right because one factor is binary. The writing is clear and the runtimes in Table I are plausible.\n\nThe remaining soft spots are secondary but real. There is no comparison to existing mobile energy storage dispatch, so the 26.27% cost reduction is not contextualized. The numerical study uses a single synthetic 14-node network with random weights and three vehicle fleets, which is thin. And Assumption 1 ignores battery state-of-charge and discharge efficiency; the authors acknowledge this, but it means the feasibility claims rest on nominal capacities.\n\nNovelty is low. This is mobile energy storage with autonomous vehicles as the motivating case, and the routing MILP is a textbook shortest-path formulation. No new mechanism distinguishes autonomous vehicles from conventional mobile storage in the model.\n\nMy recommendation: I would not send this to peer review in its current form. The routing flaw is a basic modeling error in the first component of the two-layer method, and the contribution is not novel enough to justify referee time. A desk reject with detailed comments, or a note to the authors to fix the routing constraints, add a baseline, and either model battery energy or defend the nominal-power approximation, would be appropriate. If those changes are made, the paper could become a modest but useful contribution.","headline":"The routing MILP never forces the vehicle to reach the destination, so the travel costs that feed the grid dispatch are invalid; the rest is competent but unoriginal mobile-storage work.","tokens_in":8225,"tokens_out":5054,"would_cite":false,"duration_ms":50880,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Idle autonomous vehicles can be routed and dispatched as 'mobile batteries' to cut grid operating costs, with the two-layer optimization reducing costs by 26.27% on the IEEE 14-bus system and solving 1354-bus cases in under 10 seconds.","keywords":["autonomous vehicles","power grid operations","mobile energy storage","optimal power flow","vehicle routing","McCormick relaxation","mixed-integer optimization","vehicle-to-grid"],"falsifier":"Take the reported IEEE 14-bus dispatch ($p_a=17.28$ MW, $p_b=20$ MW, $p_c=40$ MW) and check it against a battery energy constraint $E_v$ and support duration $T$ requiring $p_v T \\le E_v$; if any vehicle's required discharge energy exceeds its onboard storage, the schedule is infeasible under any model that accounts for state of charge.","tokens_in":7215,"feed_emoji":"🔋","tokens_out":9818,"duration_ms":83063,"temperature":0.7,"pith_summary":"This paper argues that idle autonomous vehicles carrying batteries can serve as a mobile, dispatchable resource for power grid operations, and that using them this way is computationally practical. The proposed method works in two layers: an offline shortest-path routing problem computes each vehicle's travel cost to every candidate bus, and a mixed-integer DC optimal power flow then decides which vehicle goes where and how much power it supplies. A McCormick relaxation replaces the bilinear product of each vehicle's power output and its binary assignment variable with four linear inequalities, and because one factor is binary the relaxation is exact rather than approximate. On the IEEE 14-bus test, integrating three vehicle fleets reduces total operating cost by 26.27%; on systems up to 1354 buses the optimization solves in under 10 seconds on a laptop. If correct, the work offers grid operators a concrete way to turn the growing autonomous vehicle fleet into flexible storage without new infrastructure.","feed_headline":"Autonomous vehicles as mobile batteries cut grid costs 26%","feed_subtitle":"A two-layer optimizer routes idle autonomous vehicles to balance grid power, solving 1,354 buses in under 10 seconds.","key_machinery":"The load-bearing object is the exact McCormick relaxation of the bilinear term in the power balance equation. For each vehicle $v$ and bus $i$, the product $y_i^v = p_v z_i^v$ is bounded by four linear inequalities, (13a)–(13d); because the assignment variable $z_i^v$ is binary, the inequalities are exact, meaning they enforce the product identity rather than a convex envelope around it. This exactness converts the nonlinear optimal power flow with vehicle routing into the mixed-integer linear program (14) that can be solved by standard solvers, and it is what makes the two-layer architecture practical: shortest-path routing (6) is solved offline to generate travel costs, and the OPF layer decides vehicle assignment and dispatch in seconds.","core_discovery":"On its own terms, the paper's central discovery is that the nonlinear coupling between vehicle power and vehicle location in the grid's power balance constraint can be handled exactly by a McCormick relaxation. For each vehicle $v$ and bus $i$, the product $y_i^v = p_v z_i^v$ is replaced by the four inequalities (13a)–(13d); because $z_i^v$ is binary, these inequalities force $y_i^v=0$ when the vehicle is not assigned to $i$ and $y_i^v=p_v$ when it is, so the reformulated problem (14) is equivalent to the original nonlinear problem (10), not an approximation. Combined with a shortest-path routing layer that supplies travel costs offline, the resulting mixed-integer program coordinates autonomous vehicles as 'mobile batteries' to maintain power balance and reduce operating cost, demonstrated by a 26.27% cost reduction on the IEEE 14-bus system and sub-10-second solve times on test systems up to 1354 buses.","pith_inferences":["A realistic battery model with state of charge and discharge efficiency would likely shrink the 26.27% saving, since some dispatched vehicles might not be able to deliver their nominal power for the full support window; a rolling-horizon extension with energy limits would quantify that shrink.","The same exact-McCormick trick applies to any optimization where a continuous variable multiplies a binary assignment variable, so the approach generalizes to siting mobile energy storage, temporary generators, or other relocatable resources in power systems.","The reported saving is tied to one synthetic 14-node transportation network with random weights; applying the method to a real city's road network with congestion and time-varying travel costs could change routes and costs, making the result a proof-of-concept rather than a guaranteed figure.","A testable extension is to compare the dispatch from the random-weight network with one from a real road network, holding the power system data fixed, to see how much the route layer affects the achievable cost reduction."],"forward_implications":["A grid operator could schedule a fleet of idle autonomous vehicles as a low-cost, driverless generation resource, with assignment and power decisions produced by a single mixed-integer program.","The sub-10-second solve times on systems up to 1354 buses indicate the method is fast enough for intra-day or near-real-time scheduling on current hardware.","The 26.27% cost reduction in the 14-bus test comes from displacing costly generators with cheaper vehicle battery power, so the potential saving grows with the share of vehicle capacity in the system.","Because the routing layer is solved offline and only travel costs enter the OPF, the plan can adapt to changing road conditions by re-solving the routing problem in real time.","The paper's runtimes assume a small number of vehicle fleets; it notes that a significantly larger fleet would require a more coordinated decision process and could increase computation time."],"supporting_citations":[{"why":"Supplies the McCormick relaxation technique that makes the bilinear product exactly reformulable as linear inequalities.","marker":"[12]"},{"why":"Provides the DC power flow model used for the grid layer in equation (9).","marker":"[10]"},{"why":"Gives the V2G technology basis underlying Assumption 1 that vehicles can inject power into the grid.","marker":"[11]"},{"why":"Motivates mobile energy storage integration and the assumption that the transportation and power networks share the same node set.","marker":"[5]"},{"why":"Reviews mobile energy storage for grid resilience and supports the shared-node network modeling assumption.","marker":"[9]"}],"fun_headline_variants":["Idle robo-cars become grid batteries, cutting costs 26%","Autonomous fleet as mobile storage: grid cost cut by 26%","Robocars as mobile batteries: grid ops get 26% cheaper","Autonomous vehicles balance grid, save 26% on costs","Grid power from parked self-driving cars: 26% cost saving"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire dispatch plan assumes every selected vehicle can reliably deliver its nominal power $p_v$ once dispatched, ignoring battery state of charge and discharging efficiency; if a dispatched vehicle is undercharged or cannot discharge at the assumed rate, the power balance at its assigned bus fails and the solution is infeasible.","fun_headline_variants_meta":{"raw":{"variants":["Idle robo-cars become grid batteries, cutting costs 26%","Autonomous fleet as mobile storage: grid cost cut by 26%","Robocars as mobile batteries: grid ops get 26% cheaper","Autonomous vehicles balance grid, save 26% on costs","Grid power from parked self-driving cars: 26% cost saving"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000546,"raw_usage":{"total_tokens":2577,"prompt_tokens":880,"completion_tokens":1697,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":496,"completion_tokens_details":{"reasoning_tokens":1603}},"tokens_in":496,"tokens_out":1697,"duration_ms":10514,"temperature":1.0,"reasoning_tokens":1603,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T20:18:41.654583+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the reported IEEE 14-bus dispatch ($p_a=17.28$ MW, $p_b=20$ MW, $p_c=40$ MW) and check it against a battery energy constraint $E_v$ and support duration $T$ requiring $p_v T \\le E_v$; if any vehicle's required discharge energy exceeds its onboard storage, the schedule is infeasible under any model that accounts for state of charge.","supporting_citations":[{"cited_title":"Solving mixed integer bilinear problems using MILP formulations,","cited_arxiv_id":null,"evidence_quote":"Supplies the McCormick relaxation technique that makes the bilinear product exactly reformulable as linear inequalities."},{"cited_title":"DC power flow revisited,","cited_arxiv_id":null,"evidence_quote":"Provides the DC power flow model used for the grid layer in equation (9)."},{"cited_title":"Distributed coordination of electric vehicles providing V2G services,","cited_arxiv_id":null,"evidence_quote":"Gives the V2G technology basis underlying Assumption 1 that vehicles can inject power into the grid."},{"cited_title":"Enhancing distribution system resilience with mobile energy storage and microgrids,","cited_arxiv_id":null,"evidence_quote":"Motivates mobile energy storage integration and the assumption that the transportation and power networks share the same node set."},{"cited_title":"Application of mobile energy storage for enhancing power grid resilience: A review,","cited_arxiv_id":null,"evidence_quote":"Reviews mobile energy storage for grid resilience and supports the shared-node network modeling assumption."}],"review_version":1}