{"id":"0f9cad08-f156-4ba6-b14d-363c4b5ab090","arxiv_id":"2412.09782","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"EI-Drive extends the CARLA simulator with latency and error models, enabling cooperative perception experiments under realistic communication conditions.","lead":"EI-Drive is a new open-source simulation platform built on CARLA that adds transmission latency and data loss to cooperative perception in autonomous driving tests. It lets researchers see how real-world network delays and errors affect vehicles that share sensor data.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The novelty claim hinges on the communication model in Section III-C, but that model is only a latency/error knob and is never validated against V2X traces or a packet-level simulator, leaving 'real-world network conditions' unsupported.","rationale":"The paper's distinctive contribution is almost entirely carried by the claim that EI-Drive models realistic communication for cooperative perception; the other components are standard CARLA, late-fusion, PID, and rule-based planning pieces. I focused on whether that claim is actually supported. Section III-C describes only a delay and a frame-loss probability, with no distributional details, no timestamp semantics, and no calibration. The experiments show qualitative degradation, but they cannot distinguish a plausible V2X impairment process from an arbitrary perturbation. I also examined Table III as an alternative concern: the scenario descriptions in Table I do not distinguish Scenarios 5-8, and the 0% success rates coexist with nonzero minimal distances, which is hard to interpret. That is a real reproducibility issue, but it attacks the experimental evidence rather than the central platform concept. The open-source release is genuine support, and it makes the proposed packet-level comparison feasible. Because the reader's conditional verdict already rests on the unvalidated communication model, my stress-test does not move the verdict.","tokens_in":12189,"tokens_out":10932,"duration_ms":121462,"concrete_test":"Re-run Coop. Scenario 1 with identical vehicle trajectories, sensor setup, and fusion code, but drive the communication model from a packet-level V2X simulator such as ns-3 with IEEE 802.11p or Veins at the same map and traffic density, instead of fixed 0.3 s / 30% knobs. Then configure EI-Drive's simple model using the mean latency and mean loss rate observed in the packet-level run, and compare the distributions of received-frame age, frame loss, success rate, and minimal distance. If the outputs match within a pre-registered tolerance (e.g., same effect sign and within 10% of the mean metrics), the abstraction is adequate and the concern does not land. If the outputs diverge materially, especially due to bursty or distance-correlated losses the simple model cannot express, the 'realistic communication model' claim is not supported as written.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that EI-Drive closes a gap in cooperative-perception simulators by incorporating realistic communication models. The load-bearing element is Section III-C's communication model, which applies a latency value and a frame-error probability to perception results. The paper does not specify the delay distribution, whether losses are independent or bursty, how timestamps are handled in the data-fusion module, or how the two parameters relate to any vehicular channel. It also presents no validation of the abstraction against real V2X measurements or a packet-level simulator. Latency and loss are necessary but not sufficient ingredients for 'real-world network conditions': real V2X latency and loss emerge from propagation, contention, packet size, and channel correlations, and for the platform's stated purpose of studying robustness under realistic impairments, the fidelity of the impairment model determines whether conclusions transfer. The experiments show that EI-Drive can degrade cooperative perception under arbitrary delay and loss, but they do not show that the chosen delay/loss process is representative. If the simple model is unrepresentative, the distinctive contribution reduces to a simulator with a delay-and-drop plugin.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents EI-Drive, an open-source simulation platform built on CARLA that integrates cooperative perception with a communication model that adds transmission latency and frame-error probability to the data exchanged between vehicles and RSUs. The paper describes the modular pipeline (sensing, perception, planning, control), the edge-AI module containing the communication model and a simple late-fusion data aggregation method, and a set of built-in scenarios. Experiments demonstrate that cooperative perception improves collision avoidance and object detection in occluded traffic scenarios, and that introducing latency and errors degrades these benefits. The stated contribution is to fill the gap left by existing cooperative-perception simulators such as OpenCDA, which do not model transmission impairments.","tokens_in":12436,"tokens_out":3347,"duration_ms":35253,"significance":"If the platform delivers on its claims, it would provide a useful open testbed for studying how communication impairments affect cooperative-perception algorithms, an area where the cited existing simulators are indeed lacking. The paper's strengths are its open-source release, modular architecture, built-in scenarios, and the use of oracle methods to isolate perception effects. There are no fitted parameters or derived predictions, so the work is not circular in the equation-derivation sense. However, the central novelty claim rests on the realism of the communication model, and that model is currently only a latency/error knob with no validation against real V2X conditions or packet-level simulation. The experimental sections also lack the statistical detail needed to support the word \"significant.\" These issues are load-bearing for the paper's main contribution but appear addressable within the scope of a revision.","major_comments":[{"comment":"The communication model is the paper's key novelty, yet it is specified only as a latency parameter and a frame-error probability applied to perception outputs. The text does not specify the delay distribution, whether losses are independent or bursty, how timestamps are handled in the fusion module, or how the parameters relate to any vehicular channel model. No validation against real V2X traces or a packet-level simulator (e.g., ns-3, Veins) is provided. Because the abstract and Section I claim evaluation under conditions that \"mirror real-world networks\" and \"real-world network conditions,\" this is load-bearing for the novelty claim. Please either add validation or a sensitivity analysis against a representative V2X model, or revise the claims to describe the platform as supporting configurable latency and error impairments rather than realistic network conditions.","section":"Section III-C and Abstract"},{"comment":"The quantitative results report means plus/minus deviations but do not state the number of trials per condition, and several entries have zero variance (e.g., Table II YOLOv5 No Coop perception 3.32 ± 0.00; Table III Coop Scenario 5 RSU Min Distance 3.03 ± 0.03; Coop Scenario 7 Vehicle success 0.00% with no uncertainty). The text claims \"significant improvements\" and \"significantly higher\" success rates without any significance tests. Please report the number of runs for each condition, specify how the error bars were computed, and perform appropriate statistical tests, or clearly mark which comparisons are deterministic single-run demonstrations rather than statistical claims.","section":"Section IV-B, Tables II and III"},{"comment":"The success rate is defined as Ncf/Ntotal, but the paper does not specify how many total attempts Ntotal were used or how the ± values in Table III were derived (e.g., across repeated episodes, bootstrap, or some other procedure). Without this information, the comparison between \"Vehicle + RSU,\" \"RSU,\" and \"Vehicle\" participation cannot be evaluated. Add the trial count and the uncertainty derivation for each entry in Table III.","section":"Section IV-B, Eq. (1)"}],"minor_comments":[{"comment":"There is a typo: \"thees environmental influences\" should be \"these environmental influences.\"","section":"Section III-B"},{"comment":"The conclusion contains \"an platform,\" which should be \"a platform.\"","section":"Section V"},{"comment":"The phrase \"EI-Drive aims to filled this gap\" should be \"EI-Drive aims to fill this gap.\"","section":"Section II-A"},{"comment":"\"Y AML file\" should be \"YAML file.\"","section":"Section III-A"},{"comment":"The text says \"An key data source\" in the data-fusion paragraph; this should be \"A key data source.\"","section":"Section III-C"},{"comment":"The descriptions for Coop Scenario 5&6 and Coop Scenario 7&8 are identical; please clarify the difference between these scenarios (e.g., which agents are present or what traffic conditions differ).","section":"Table I"},{"comment":"The x-axis label \"Ticks\" is unclear; please specify the unit (e.g., simulation time steps) and state whether the plotted curves are averages over multiple episodes with error bars.","section":"Figure 6"},{"comment":"The communication model description says latency can be \"deterministic or stochastic\" but gives no details on the stochastic distributions; a brief description of supported distributions would help users understand the model's flexibility.","section":"Section III-C"}],"recommendation":"major_revision","confidential_remarks":"The paper is a systems/demonstration contribution rather than a new algorithm. Its fit depends on whether the venue values open-source simulation platforms with modular communication-aware cooperative perception. The main risk is that the communication model's simplicity is not adequately validated, which directly affects the paper's central \"realistic\" claim. The experimental presentation also needs to meet standard statistical reporting expectations before the quantitative conclusions can be assessed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main thing you should know: EI-Drive is a real, usable open-source simulator. It puts a cooperative perception stack (late fusion of YOLOv5/SSD detections, plus an oracle mode) into CARLA and lets you inject a fixed latency and a frame-error probability into the perception results before they reach the ego vehicle. The surrounding machinery—A* planning, PID control, scenario definitions, spectator vehicles and RSUs—is standard but cleanly modular. If you need to see how cooperative perception degrades under stale or missing bounding boxes, this platform will save you weeks.\n\nThe paper does some things well. The hidden-vehicle intersection with an occluding firetruck is a good stress case, and the RSU-versus-vehicle comparison in heavy traffic is a nice demonstration of why mounting height matters. Offering oracle perception alongside the learned detectors is good practice; it lets users separate detector noise from fusion effects.\n\nThe soft spots are real but not fatal. The abstract and intro claim 'realistic communication models,' but what is implemented is a two-knob delay/loss model: a scalar latency (deterministic or stochastic, distribution unspecified) and an independent frame-loss probability. There is no validation against V2X traces or packet-level simulators, no modeling of burstiness or channel correlation. So 'realistic' is doing more work than the evidence supports. The authors should either calibrate the model or soften the language to 'simple transmission impairment model.'\n\nThe experimental reporting is also thin. Table II has zero-variance entries (3.32 ± 0.00, 3.91 ± 0.00), no number of trials is given anywhere, and the word 'significant' appears without any statistical test. The qualitative conclusions—cooperation helps, latency and errors hurt—are solid, but the precision implied by the ± values is unsupported. Fixing this is easy: report trial counts, run more seeds, and drop the significance language unless a test is actually run.\n\nThe novelty claim is credible but modest. OpenCDA has cooperative perception without latency/error; ns-3/Veins have network fidelity without the driving stack. EI-Drive sits in between, which is a legitimate gap. The paper says 'to our best knowledge,' which is appropriately hedged.\n\nJudged as a platform paper, this is legitimate and useful. The central argument holds: you can study transmission impairments on cooperative perception in this simulator. The unvalidated communication model is the main weakness, and it is addressable in revision. I would send this to peer review with a request for major revision: calibrate or rebrand the communication model, add statistical details, and tighten the claims. A serious referee will find value in it.","headline":"A genuinely useful open-source cooperative-perception simulator, but the 'realistic communication' claim outruns the two-knob delay/loss model it actually ships.","tokens_in":12921,"tokens_out":2558,"would_cite":true,"duration_ms":26542,"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":"EI-Drive is a simulation platform that adds transmission latency and frame errors to cooperative perception, and its experiments show these impairments significantly degrade autonomous driving safety.","keywords":["autonomous driving","cooperative perception","vehicular communication","simulation platform","transmission latency","frame error","data fusion","CARLA"],"falsifier":"A direct comparison of EI-Drive's delivered bounding-box delays and loss patterns against a packet-level V2X simulation, such as ns-3 or Veins with a realistic 5.9 GHz channel model, or against a field-collected V2X trace, would settle whether the simple latency-plus-error model reproduces the temporal correlation and burstiness of real wireless delivery; if real delays are correlated across consecutive frames while EI-Drive treats them as independent, the platform's predictions about which algorithms are robust would likely be wrong.","tokens_in":12020,"feed_emoji":"🚗","tokens_out":7622,"duration_ms":74489,"temperature":0.7,"pith_summary":"Cooperative perception lets vehicles and roadside units share what they see, but real wireless links add delay and drop data; most simulation platforms that test cooperative perception pretend the shared data arrives instantly and intact. This paper introduces EI-Drive, a CARLA-based open simulation platform that adds a communication model placing transmission latency and frame-error probability on the perception results shared between agents. The authors' experiments with collision-avoidance and object-detection scenarios show that cooperative perception improves safety margins and detection counts when the network is perfect, that the added latency and errors measurably shrink those gains, and that data from elevated roadside units holds up better than data from other vehicles in heavy traffic. The point is to give the autonomous-driving community a way to evaluate and train algorithms under network conditions closer to what real vehicles face.","feed_headline":"Latency and frame errors degrade cooperative perception safety","feed_subtitle":"A CARLA-based open platform lets AV researchers test perception under transmission latency and errors.","key_machinery":"The load-bearing component is the communication model inside the edge-AI module: a delay-and-loss filter applied to every shared perception message. It treats transmission latency as a time offset, set to be either deterministic or stochastic, and transmission error as an adjustable frame-drop probability, with data fusion then averaging overlapping bounding boxes from the messages that survive. This small mechanism is what converts a standard CARLA cooperative-driving pipeline into a tool for studying how network conditions affect perception and planning.","core_discovery":"The paper claims that EI-Drive is the first open simulation platform to combine cooperative perception with an explicit model of wireless transmission latency and frame errors, and that this combination changes conclusions about autonomous driving safety. Built on CARLA, the platform intercepts the bounding-box perception results that cooperating vehicles and RSUs send to an ego vehicle, applies a configurable delay and a probability of frame loss to each message, and fuses the surviving boxes with a simple late-fusion average. In the reported experiments a 0.3 second latency and a 30% error rate significantly reduce the minimal distance maintained during collision avoidance, and scenarios without roadside units drop to success rates as low as 0% in heavy traffic, while RSU-assisted fusion keeps success near 100%. The paper's claim is not that any particular detector is better, but that network impairments are first-order factors in cooperative-driving performance and that an open platform allowing such impairments to be configured and studied is needed.","pith_inferences":["The independent per-frame error model likely understates real V2X conditions, where losses arrive in bursts and delay is correlated with congestion; coupling EI-Drive with a packet-level network simulator would reveal whether the robustness conclusions hold.","The 0.3 s / 30% parameter choice is a single operating point; sweeping error rate and latency continuously could expose a phase transition where cooperative perception stops beating single-vehicle perception, which would give system designers a concrete link-budget target.","Because the same delay-and-drop filter can be applied to the ego's own perception pipeline, the communication module doubles as a way to test robustness to sensor processing delay and sensor dropout without any network at all.","The reported vehicle-only failures in scenarios 7 and 8 hint that in dense traffic, cooperation from occluded participants can be misleading; testing with noisy detectors instead of the oracle might show earlier points where cooperation turns harmful."],"forward_implications":["Cooperative perception with perfect communication improves collision-avoidance safety, shown by larger minimum distances and higher success rates in every tested scenario.","Imposing 0.3 s latency or a 30% frame-error rate degrades those gains, so benchmarks that omit communication effects overstate how safe a cooperative algorithm is.","RSUs with an elevated, unobstructed view sustain near-100% success in heavy traffic, whereas relying on a spectator vehicle alone can drop to 0% success when occlusion is severe.","The platform's modular design lets researchers swap in custom fusion methods, detectors, communication parameters, and scenario definitions to test robustness under network stress."],"supporting_citations":[{"why":"Provides the CARLA urban driving simulation environment that EI-Drive builds on.","marker":"[6]"},{"why":"OpenCDA is the existing cooperative driving platform that, the paper argues, lacks transmission latency and error features; it anchors the gap EI-Drive fills.","marker":"[20]"},{"why":"Supplies evidence that low-latency and ultra-reliable V2V communication is a practical challenge, motivating the need to model latency.","marker":"[4]"},{"why":"Connects communication and sensing impairments to driving performance, used to justify why latency and errors belong in a driving simulator.","marker":"[5]"},{"why":"ns-3 is a representative packet-level network simulator whose C++ implementation the paper says is difficult to integrate with Python-based AV simulators.","marker":"[21]"},{"why":"OMNeT++ is an extensible communication simulator also cited as hard to interface with mainstream Python driving simulators.","marker":"[22]"},{"why":"Veins is a V2X simulation framework built on OMNeT++, illustrating the integration gap EI-Drive aims to close.","marker":"[23]"},{"why":"Simu5G is another OMNeT++-based simulator for 5G networks, further showing that current network tools do not plug into AV platforms.","marker":"[24]"},{"why":"VSimRTI is a V2X simulation runtime that the paper says does not involve cooperative perception, marking a different limitation of prior tools.","marker":"[25]"}],"fun_headline_variants":["EI-Drive simulates packet loss and delay to expose AV safety risks","Latency and errors in V2X comms reduce cooperative safety in tests","New open simulator shows network delays undermine self-driving cooperation","Realistic comm models in simulation reveal safety drops for connected AVs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The communication model — a fixed or randomly drawn delay plus an independent frame-drop probability — is assumed to faithfully represent real vehicular network conditions, but the paper does not validate this model against real V2X trace data or a packet-level network simulator.","fun_headline_variants_meta":{"raw":{"variants":["EI-Drive simulates packet loss and delay to expose AV safety risks","Latency and errors in V2X comms reduce cooperative safety in tests","New open simulator shows network delays undermine self-driving cooperation","Realistic comm models in simulation reveal safety drops for connected AVs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000458,"raw_usage":{"total_tokens":2280,"prompt_tokens":912,"completion_tokens":1368,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":528,"completion_tokens_details":{"reasoning_tokens":1293}},"tokens_in":528,"tokens_out":1368,"duration_ms":11970,"temperature":1.0,"reasoning_tokens":1293,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T16:43:00.981148+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct comparison of EI-Drive's delivered bounding-box delays and loss patterns against a packet-level V2X simulation, such as ns-3 or Veins with a realistic 5.9 GHz channel model, or against a field-collected V2X trace, would settle whether the simple latency-plus-error model reproduces the temporal correlation and burstiness of real wireless delivery; if real delays are correlated across consecutive frames while EI-Drive treats them as independent, the platform's predictions about which algorithms are robust would likely be wrong.","supporting_citations":[{"cited_title":"Towards low-latency and ultra-reliable vehicle-to-vehicle communication,","cited_arxiv_id":null,"evidence_quote":"Supplies evidence that low-latency and ultra-reliable V2V communication is a practical challenge, motivating the need to model latency."},{"cited_title":"Impact of sensing errors on headway design: From alpha-fair group safety to traffic throughput,","cited_arxiv_id":null,"evidence_quote":"Connects communication and sensing impairments to driving performance, used to justify why latency and errors belong in a driving simulator."},{"cited_title":"An overview of the omnet++ simulation environment,","cited_arxiv_id":null,"evidence_quote":"OMNeT++ is an extensible communication simulator also cited as hard to interface with mainstream Python driving simulators."},{"cited_title":"Simu5g–an omnet++ library for end-to-end performance evaluation of 5g networks,","cited_arxiv_id":null,"evidence_quote":"Simu5G is another OMNeT++-based simulator for 5G networks, further showing that current network tools do not plug into AV platforms."},{"cited_title":"V2x simulation runtime infrastructure vsimrti: An assessment tool to design smart traffic management systems,","cited_arxiv_id":null,"evidence_quote":"VSimRTI is a V2X simulation runtime that the paper says does not involve cooperative perception, marking a different limitation of prior tools."}],"review_version":1}