{"id":"6b43d8d6-155d-4691-8de7-dcd684e1f054","arxiv_id":"2506.19781","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A wheeled robot with a Starlink terminal and multi-modal sensors produces the first synchronized dataset linking mobile satellite communication performance to motion and environmental occlusion.","lead":"The authors mounted a Starlink Mini terminal on a wheeled robot with cameras, LiDAR, and IMU, and released a synchronized dataset of satellite link quality, motion, and sky visibility. The dataset lets researchers study how movement and obstacles affect mobile satellite internet, an increasingly important question as phones and vehicles gain satellite connectivity.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 3.2's unvalidated 'sub-millisecond alignment' and LEOViz's gRPC parsing are the load-bearing hinge: without ground-truth timestamp/parser checks, the synchronized dataset claim is unverifiable.","rationale":"The reader's weakest assumption aligns with the load-bearing concern. The central novelty is a synchronized multi-modal dataset; if the synchronization or the parsed Starlink metrics are wrong, then the dataset is not a reliable community resource, and the correlations in Section 5 (velocity vs. latency, occlusion vs. RTT) are not trustworthy. The concern is not that the team's hardware is implausible—the Unitree GO2 + Starlink Mini + Livox Mid-360 + fisheye setup is credible, and the use of LEOViz is a reasonable choice. Credit is due for building the platform and for attempting to release data. However, the paper provides no synchronization validation, no parser validation, no error bars, and no access instructions for the data. These are exactly the artifacts a dataset paper must supply. The reader's CONDITIONAL verdict is appropriate: the claim should be accepted only after the authors publish validation traces and dataset access. I did not find a more damaging internal inconsistency: the hardware description and the analysis pipeline are coherent, and the preliminary findings are plausible but not load-bearing on their own. One caveat: if the dataset is not actually released, the central claim would fail entirely, but that is a verification matter rather than an argument flaw, and the GitHub link suggests an intent to release.","tokens_in":8023,"tokens_out":4904,"duration_ms":53483,"concrete_test":"Use the released ROS bags to perform an event-lag validation on one tree-cover session. Threshold the upward fisheye's sky-visibility fraction to detect the moment the robot enters canopy occlusion and threshold the Starlink obstruction state/SNR series for the corresponding signal event. Compute the cross-correlation peak lag and its variance over repeated crossings. If the median lag exceeds one Starlink sample period (1 s) or the lag variance is comparable to 1 s, the claimed sub-millisecond alignment cannot hold for the communication stream. Cross-check by comparing LEOViz-parsed 1 Hz throughput/RTT against the independently recorded 10 Hz ICMP probes in the same session; systematic offsets would indicate a parser or timestamp problem.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is not the hardware alone but the release of a synchronized multi-modal dataset linking Starlink metrics to motion and sky visibility. The dataset's scientific value therefore turns on two links, and neither is evidenced in the preprint. First, Section 3.2 asserts a dedicated synchronization node 'achieving sub-millisecond alignment accuracy across all modalities.' No method, error analysis, or validation trace is provided. This is especially suspect for the Starlink stream: the terminal is accessed only through LEOViz's parsing of Starlink's gRPC status interface (Section 3.2), which supplies 1 Hz statistics; those statistics arrive as software-timestamped gRPC messages, not hardware-triggered samples. A sub-millisecond alignment claim for a 1 Hz stream requires knowing the terminal's internal measurement epoch, which the paper never establishes. Second, LEOViz's parsed values (throughput, RTT, SNR, obstruction state) are used as ground truth throughout Section 5, but no validation against independent measurements is shown. If LEOViz mis-parses a field, or if the terminal-reported statistics are delayed relative to the sensor streams, every cross-modal correlation in the dataset is compromised. This is a correctness risk, not a novelty dispute: the preliminary findings in Section 5 would remain internally consistent while being systematically misaligned. The paper's own Section 4 promises 'Metadata: ... time synchronization information' but the preprint contains no such calibration data.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces the Starlink Robot, a wheeled Unitree GO2 platform equipped with a Starlink Mini terminal, an upward-facing fisheye camera, a Livox Mid-360 LiDAR, IMU, and onboard computing. The authors describe the hardware and software architecture, a multi-modal dataset of communication metrics, motion data, sky visibility, and 3D environmental context, and a preliminary analysis of how movement speed and environmental occlusion affect Starlink latency. Satellite state and terminal statistics are collected via LEOViz's parsing of Starlink's gRPC interface. The central claims are that this is the first mobile robotic platform for Starlink measurement and that the released synchronized dataset enables future mobile satellite communication research. The paper is written as a platform-and-dataset contribution, with descriptive rather than hypothesis-driven results.","tokens_in":8431,"tokens_out":2860,"duration_ms":32484,"significance":"If the synchronization and the parsed Starlink metrics are validated, the dataset would be a valuable community resource: it is, to my knowledge, one of the first publicly described mobile Starlink measurement platforms with co-registered motion, sky-view, and communication data. The paper ships an open GitHub repository, a clear system architecture diagram, and descriptive observations about latency behavior under motion and tree cover. The strengths are the reproducible platform design and the breadth of modalities. However, the significance is conditional on two unverified links: the claimed sub-millisecond alignment of all streams, and the correctness of LEOViz-parsed terminal statistics as ground truth. The preliminary analysis currently provides qualitative pattern reports rather than quantitative evidence, which limits the scientific contribution in its present form.","major_comments":[{"comment":"The claim of 'sub-millisecond alignment accuracy across all modalities' is load-bearing for the synchronized dataset claim but is unsupported by the manuscript. No synchronization method, clock model, timestamp-correction algorithm, or validation trace is provided. For the Starlink 1 Hz stream, which arrives through LEOViz parsing of software-timestamped gRPC messages, sub-millisecond alignment requires knowing the terminal's internal measurement epoch, which the paper never establishes. Please add an alignment validation, for example residual clock offsets, aligned event timestamps, or a hardware-triggered cross-check between sensors and the Starlink stream.","section":"3.2 Software Architecture"},{"comment":"LEOViz-parsed terminal statistics (RTT, SNR, throughput, obstruction state) are treated as ground truth throughout Section 5, but no independent validation is shown. A systematic parsing error or a software-timestamp delay in the gRPC data would compromise every cross-modal correlation while leaving the reported time series internally consistent. Please validate the parsed fields against independent measurements (e.g., concurrent ICMP RTT probes, external throughput tests, or manual obstruction events) and report per-field error statistics or timing offsets.","section":"4 Dataset Description and 5 Preliminary Analysis"},{"comment":"The central findings—'minimal impact' of velocity on RTT, 'predictable 15-second handover intervals', and 'substantial instability' in tree-covered environments—are based on qualitative inspection of a small number of time-series figures (Figures 9-12 and 15) with no error bars, no session counts, no per-condition summary statistics, and no statistical tests. These claims outrun the evidence as presented. Please report medians/quantiles per speed and environment condition, the number of experimental runs, and, where appropriate, inferential tests or effect sizes.","section":"5.1 and 5.2 Preliminary Analysis"},{"comment":"Dataset availability is not verifiable from the manuscript. The abstract points to a general GitHub repository, but Section 4's promise of 'download' and 'Metadata: time synchronization information' is not backed by a persistent dataset identifier, a download URL, a schema description, checksums, or a concrete list of the metadata contents. The dataset's value is the central contribution, so the paper should include a stable dataset link and a complete metadata table.","section":"4 Dataset Description and Abstract"}],"minor_comments":[{"comment":"The phrase 'first mobile robotic platform' should be softened or carefully scoped, since reference [9] already reports vehicle-mounted Starlink measurements; please clarify the distinction (e.g., robot-centric, multi-modal, synchronized, openly released).","section":"1 Introduction"},{"comment":"Please add axis labels, units, and legends to the RTT figures, and make the handover step changes and the 1 Hz sample points legible; the current figures are hard to read at the printed scale.","section":"Figures 9, 10, 12, 15"},{"comment":"Minor formatting issue: in the 'Data Format and Organization' paragraph, 'GPS coordinates;CSV files' is missing a space after the semicolon.","section":"4 Dataset Description"},{"comment":"Please clarify the relationship between LEOViz data and the Starlink gRPC interface, and state explicitly which fields are parsed by LEOViz and which are recorded directly by the robot; this distinction matters for reproducibility.","section":"3.2 Software Architecture"}],"recommendation":"major_revision","confidential_remarks":"The dataset availability is the crux of this submission. If the synchronized dataset and the parsing/validation code are not made available with a persistent identifier and concrete access instructions before publication, the contribution reduces to a hardware description plus anecdotal measurements, which would fall below the bar for this venue. I would also check that the LEOViz-derived terminal statistics can be redistributed under LEOViz's license terms, since the dataset's core value depends on those fields."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The Starlink Robot paper is worth taking seriously. The core contribution is a first integration of a Starlink Mini on a wheeled robot with synchronized overhead camera, LiDAR, IMU, and communication metrics, plus a dataset covering open and tree-lined environments. That is a real gap: prior mobile Starlink measurements lacked precisely the synchronized motion and sky-visibility context this platform captures. The hardware choices are sensible, and releasing both design and data is a community service. If the dataset is sound, it enables work on motion-aware protocols and outage prediction that nobody else has done with public data.\n\nThe weak point is exactly where the reader's stress-test lands. Section 3.2 claims 'sub-millisecond alignment accuracy across all modalities' without showing method, error analysis, or validation traces. For a 1 Hz gRPC-parsed stream from the Starlink terminal, that claim is not self-evident; the terminal's internal measurement epoch is unknown, and software timestamps are not hardware triggers. The paper also uses LEOViz's parsed throughput, RTT, SNR, and obstruction state as ground truth throughout Section 5 without any independent check against a known-good measurement. If LEOViz mis-parses a field or the terminal's stats are delayed relative to the sensor streams, every cross-modal correlation in the dataset is compromised. The paper's own Section 4 promises 'time synchronization information' in the metadata, but the preprint contains none.\n\nThe preliminary analysis is also thin: no error bars, no statistical tests, just qualitative descriptions of RTT traces. The observed patterns are plausible, and the open-vs-tree comparison is suggestive, but it is not yet evidence. These are addressable issues, not fatal ones. The platform itself could still be cited, and the dataset could be genuinely valuable after validation.\n\nI agree with the reader's CONDITIONAL verdict and think the stress-test note holds up on reading the paper. The paper deserves peer review, not desk rejection. A serious referee should ask for three things: (1) validation of the sub-millisecond synchronization claim with concrete timestamps and offsets, (2) an independent sanity check of LEOViz fields against direct measurements (e.g., known throughput from iperf or RTT from a separate tool), and (3) basic statistics (means, variances, sample counts) for the Section 5 comparisons. With those, this becomes a strong dataset paper and a useful community resource.","headline":"A genuinely useful platform/dataset paper whose load-bearing synchronization and LEOViz-parsing claims need validation before the dataset is trusted as ground truth.","tokens_in":8812,"tokens_out":1475,"would_cite":true,"duration_ms":18149,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A wheeled robot carrying a Starlink terminal and a synchronized sensor suite provides the first platform and dataset for studying how motion and sky occlusion affect satellite internet, with preliminary results showing tree cover—not…","keywords":["mobile satellite communication","Starlink","robotic platform","multimodal dataset","LEO satellites","sky visibility","time synchronization","motion-aware connectivity"],"falsifier":"Compare the aligned dataset against an independent ground-truth time source: place a bright LED flash in the fisheye view synchronized to a GPS pulse while simultaneously logging the Starlink terminal's gRPC status; if the reported timestamps of the flash and the communication metrics disagree beyond the stated alignment error, or if LEOViz's parsed 1 Hz metrics diverge from a direct terminal query run in parallel, the central claim of a synchronized multi-modal dataset loses its basis.","tokens_in":7851,"feed_emoji":"📡","tokens_out":4176,"duration_ms":40350,"temperature":0.7,"pith_summary":"The paper introduces a wheeled robot carrying a Starlink satellite terminal, an upward-facing fisheye camera, LiDAR, and IMU, and argues this is the first platform able to measure satellite internet performance during controlled movement with full environmental context. Its central goal is to provide the research community with an open, synchronized multi-modal dataset—communication metrics, satellite geometry, motion, sky visibility, and 3D surroundings—so that mobile satellite connectivity can be studied where static measurements fall short. Using this setup, the authors report preliminary findings that pedestrian-scale speed barely affects Starlink link quality, while tree cover and visibility constraints cause frequent RTT spikes and unstable handovers. A sympathetic reader would see the contribution as the reproducible measurement infrastructure itself, plus early evidence that occlusion, not velocity, dominates the mobile experience.","feed_headline":"Trees, not speed, break mobile Starlink links","feed_subtitle":"A new robot-mounted Starlink dataset pairs satellite metrics with sky and motion data to explain dropouts.","key_machinery":"The central object is the integrated platform: a Unitree GO2 wheeled robot with a Starlink Mini terminal, an upward-facing 185-degree fisheye camera, a Livox Mid-360 LiDAR, and IMU plus wheel odometry, all time-synchronized through a ROS-based core. The named software tool LEOViz—an open-source parser for Starlink's gRPC status interface—supplies the 1 Hz satellite positions, elevation and azimuth angles, SNR, and connection status. The mechanism that carries the argument is the synchronization pipeline that aligns high-rate sensor streams (camera about 15 Hz, LiDAR about 25 Hz, IMU about 8 Hz) with 1 Hz Starlink statistics and 10 Hz ICMP probes into unified HDF5 files, making it possible to correlate sky visibility and motion with link quality.","core_discovery":"The central claim is that a purpose-built mobile robotic platform with a Starlink Mini terminal and synchronized sensors can capture, for the first time, the full physical context of satellite communication while moving, and that the resulting dataset reveals patterns invisible to stationary terminals. The paper reports two preliminary findings: at slow (about 0.8 m/s) and fast (about 2.0 m/s) pedestrian speeds, RTT remains concentrated in the 35–45 ms range with handovers every roughly 15 seconds, so velocity itself has minimal impact; in tree-covered environments, RTT becomes unstable with spikes of 40–100 ms, showing that sky visibility and handover options, not motion speed, drive mobile satellite performance. The authors also claim the platform and dataset enable motion-aware protocols, connectivity-disruption prediction, and communication-aware path planning.","pith_inferences":["An obvious extension the paper leaves implicit is using the fisheye sky-visibility stream alone as a proxy for link quality, letting mobile devices with simple cameras anticipate dropouts without relying on the terminal's internal telemetry.","If the sub-millisecond synchronization claim holds, the dataset supports supervised learning that maps 3D LiDAR geometry directly to expected RTT variance, a step toward communication-aware navigation in cluttered cities.","The reported 1 Hz terminal statistics may smooth out fast handover transients, so a testable follow-up would re-run the same routes while logging directly from the terminal's gRPC stream at a higher rate rather than through LEOViz."],"forward_implications":["At pedestrian speeds, Starlink's phased-array beam steering compensates for motion, so delivery robots and walking users need not treat velocity as a primary link risk.","Environmental occlusion dominates mobile satellite performance, so communication-aware path planning and predictive handover should focus on sky visibility and canopy gaps.","Researchers can use the released dataset to train models that predict RTT degradation or handover timing from fisheye sky images and LiDAR geometry.","The roughly 15-second handover cadence manifests as predictable RTT step changes, enabling disruption prediction in open environments.","The platform's open design allows other groups to replicate the measurement setup and extend it to new terminal models and constellations."],"supporting_citations":[{"why":"Supplies the central data-collection mechanism: LEOViz parses Starlink's gRPC status interface and records satellite positions, elevation/azimuth, signal quality, and connection status at 1 Hz.","marker":"[3, 23]"},{"why":"The main prior mobility baseline the paper contrasts with: vehicle-mounted Starlink tests that reported increased latency variance but lacked synchronized motion and environmental context.","marker":"[9]"},{"why":"Establishes static Starlink throughput and latency baselines across geographic locations, which the mobile measurements extend and compare against.","marker":"[8]"},{"why":"Provides the weather-impact baseline for Starlink links, showing rain fade is milder than for geostationary satellites and framing environmental influence on link quality.","marker":"[16]"},{"why":"A prior LEO satellite network measurement dataset that motivates the dataset contribution and positions the new multi-modal mobile dataset within the measurement community.","marker":"[21]"},{"why":"Documents the tradition of robot-mounted wireless datasets, supporting the argument that controlled mobility reveals network behavior invisible in static deployments.","marker":"[1]"},{"why":"Shows how aerial platforms have been used to characterize 5G coverage, serving as the robotics-side precedent the platform adapts to satellite communication.","marker":"[14]"}],"fun_headline_variants":["Starlink robot shows trees, not speed, cause dropouts","Foliage, not motion, determines mobile Starlink reliability","New robot dataset: tree cover breaks Starlink, speed doesn't","Speed isn't the culprit—tree cover kills moving Starlink","Starlink on a robot: trees disrupt links, speed does not"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The dataset's usefulness rests on the claim that all streams—including the 1 Hz Starlink statistics parsed by LEOViz—are time-aligned with sub-millisecond accuracy, but the paper provides no validation data or error analysis for that alignment.","fun_headline_variants_meta":{"raw":{"variants":["Starlink robot shows trees, not speed, cause dropouts","Foliage, not motion, determines mobile Starlink reliability","New robot dataset: tree cover breaks Starlink, speed doesn't","Speed isn't the culprit—tree cover kills moving Starlink","Starlink on a robot: trees disrupt links, speed does not"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001008,"raw_usage":{"total_tokens":4217,"prompt_tokens":861,"completion_tokens":3356,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":477,"completion_tokens_details":{"reasoning_tokens":3263}},"tokens_in":477,"tokens_out":3356,"duration_ms":25458,"temperature":1.0,"reasoning_tokens":3263,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:23:59.784216+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the aligned dataset against an independent ground-truth time source: place a bright LED flash in the fisheye view synchronized to a GPS pulse while simultaneously logging the Starlink terminal's gRPC status; if the reported timestamps of the flash and the communication metrics disagree beyond the stated alignment error, or if LEOViz's parsed 1 Hz metrics diverge from a direct terminal query run in parallel, the central claim of a synchronized multi-modal dataset loses its basis.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The main prior mobility baseline the paper contrasts with: vehicle-mounted Starlink tests that reported increased latency variance but lacked synchronized motion and environmental context."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes static Starlink throughput and latency baselines across geographic locations, which the mobile measurements extend and compare against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"A prior LEO satellite network measurement dataset that motivates the dataset contribution and positions the new multi-modal mobile dataset within the measurement community."},{"cited_title":"CRAWDAD: Community Resource for Archiving Wireless Data at Dartmouth","cited_arxiv_id":null,"evidence_quote":"Documents the tradition of robot-mounted wireless datasets, supporting the argument that controlled mobility reveals network behavior invisible in static deployments."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows how aerial platforms have been used to characterize 5G coverage, serving as the robotics-side precedent the platform adapts to satellite communication."}],"review_version":2}