{"id":"e99defdb-02e1-46e9-89a4-ff0905f00030","arxiv_id":"2607.02037","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"An RL policy trained in simulation with randomized vessel dynamics and conditioned on history via a teacher-student latent module achieves zero-shot real deployment on two ASVs, outperforming non-adaptive baselines by up to 58% in position error.","lead":"The paper presents an adaptive reinforcement learning controller for autonomous surface vehicles that uses a teacher-student architecture to infer latent dynamics from interaction history, enabling zero-shot deployment across different real-world platforms. A smart generalist might read it to see how machine learning can reduce the engineering burden of customizing controllers for varied hardware in robotics applications.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Randomization in simple analytical dynamics model may not span real hydrodynamic variability for zero-shot transfer","rationale":"The reader's weakest assumption directly identifies the sim-to-real gap with the analytical model as the load-bearing point. Full-text details on randomization ranges, latent dimension, and any sim-to-real validation would be needed to assess it further, but the core risk remains unaddressed by the reported experiments alone. This keeps the verdict at UNVERDICTED rather than moving it to ACCEPT or REJECT.","tokens_in":1652,"tokens_out":365,"duration_ms":26995,"concrete_test":"Retrain the policy using the same teacher-student architecture but with dynamics randomization drawn from a high-fidelity hydrodynamic simulator (e.g., including full 3D added-mass and damping matrices fitted to real platform data) instead of the analytical model; deploy both versions zero-shot to the two real platforms and check whether the analytical-model version retains within 10% of the reported MAE improvement over non-adaptive baselines.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that a policy trained under parameter randomization of a basic analytical vessel model (with teacher-student latent inference from history) will generalize to unmodeled real effects such as nonlinear damping, wave-induced forces, and platform-specific hydrodynamics on two unseen vessels. This assumption is least secure because the training distribution is generated from the analytical equations rather than a high-fidelity simulator; if the randomization ranges do not cover the support of real dynamics (or if the latent module overfits to sim artifacts), the reported 58% MAE improvement and near-parity with tuned controllers could fail to hold under broader testing. The paper's success on two platforms does not by itself confirm coverage of the real distribution.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper claims to introduce an adaptive RL policy for ASV trajectory tracking that achieves zero-shot cross-platform deployment. It conditions the policy on interaction history via a teacher-student architecture that infers a latent representation of unknown vessel dynamics. Training occurs in simulation by randomizing parameters of a simple analytical dynamics model; the resulting policy is deployed without fine-tuning on two distinct real-world platforms. Experiments report that the adaptive policy reduces position mean absolute error by up to 58% relative to non-adaptive learning baselines while approaching the accuracy of a platform-specific tuned controller.","tokens_in":1788,"tokens_out":515,"duration_ms":43288,"significance":"If substantiated with adequate experimental detail, the result would demonstrate that standard partial-observability techniques combined with dynamics randomization in a low-fidelity analytical model can produce policies that transfer to real, unmodeled hydrodynamics across platforms. This is noteworthy because it avoids both per-platform retuning and high-fidelity simulators, directly addressing a practical barrier in ASV deployment. The real-world validation on two platforms supplies concrete evidence of transfer, which strengthens the contribution relative to purely simulated studies.","major_comments":[{"comment":"Abstract and §5 (Real-world Experiments): the central zero-shot claim rests on quantitative gains (up to 58% MAE reduction) yet supplies no training hyperparameters, randomization ranges for the analytical model parameters, number of trials, error bars, statistical tests, or exclusion criteria. These omissions are load-bearing because they prevent independent verification that the reported improvement is robust rather than an artifact of a narrow test set or favorable randomization.","section":"Abstract and §5"},{"comment":"§4 (Simulation Training): the method randomizes vessel dynamics inside a simple analytical model to produce a policy that generalizes to real hydrodynamic effects on unseen platforms. No explicit ranges, sampling distributions, or coverage analysis for parameters such as damping or added mass are provided. This assumption is load-bearing for the cross-platform claim; if the randomization support does not intersect the distribution of real nonlinear damping and wave-induced forces, the observed transfer on two platforms does not establish broader generalization.","section":"§4"}],"minor_comments":[{"comment":"Notation for the latent state inferred by the teacher module is introduced without an explicit equation linking it to the policy input; a single clarifying equation in §3 would improve readability.","section":"§3"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"Thank you for the constructive review. The comments correctly identify omissions that affect reproducibility and the strength of the generalization claim. We address each point below and will revise the manuscript to supply the requested details.","responses":[{"response":"We agree these details are required for independent verification. In the revised manuscript we will add, in §5 and a new appendix, the full training hyperparameters, the exact randomization ranges and sampling distributions for all analytical-model parameters, the number of real-world trials per platform, error bars (standard deviation across trials), results of paired statistical tests on the MAE differences, and any trial-exclusion criteria that were applied. The 58 % figure will be reported with these supporting statistics.","revision_made":"yes","referee_comment":"[Abstract and §5] Abstract and §5 (Real-world Experiments): the central zero-shot claim rests on quantitative gains (up to 58% MAE reduction) yet supplies no training hyperparameters, randomization ranges for the analytical model parameters, number of trials, error bars, statistical tests, or exclusion criteria. These omissions are load-bearing because they prevent independent verification that the reported improvement is robust rather than an artifact of a narrow test set or favorable randomization."},{"response":"We accept that explicit randomization details and coverage analysis are necessary. Section 4 will be expanded to list the precise ranges and sampling distributions (uniform over intervals) for every randomized parameter, including linear and quadratic damping coefficients and added-mass terms. We will also add a short coverage discussion that compares the simulated parameter support against typical literature values for the two real platforms; any gaps relative to unmodeled nonlinear or wave effects will be noted as a limitation.","revision_made":"yes","referee_comment":"[§4] §4 (Simulation Training): the method randomizes vessel dynamics inside a simple analytical model to produce a policy that generalizes to real hydrodynamic effects on unseen platforms. No explicit ranges, sampling distributions, or coverage analysis for parameters such as damping or added mass are provided. This assumption is load-bearing for the cross-platform claim; if the randomization support does not intersect the distribution of real nonlinear damping and wave-induced forces, the observed transfer on two platforms does not establish broader generalization."}],"tokens_in":1399,"tokens_out":479,"duration_ms":34510,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing to know is that a single policy trained under randomized simple vessel dynamics plus a teacher-student module for inferring latent platform properties from history tracks trajectories on two different real ASVs without any fine-tuning, cutting position MAE by up to 58% versus non-adaptive RL baselines and getting close to a platform-specific tuned controller.\n\nWhat is new is the concrete zero-shot cross-platform result on actual marine hardware using the standard partial-observability approach of conditioning on interaction history. The paper does well by reporting real deployments on two vessels rather than stopping at simulation, which gives the claim some grounding.\n\nThe soft spot is the training distribution. Randomizing parameters inside a basic analytical dynamics model may not span the real hydrodynamic effects (nonlinear damping, wave forces, platform-specific interactions) that appear on unseen vessels. The two successful platforms are encouraging, but they do not by themselves confirm that the randomization covered the relevant support; if the latent module is mostly fitting sim artifacts, broader testing could show larger gaps. The abstract also omits training details, error bars, statistical tests, and how the randomization ranges were chosen, which makes it harder to judge how much of the gain comes from the method versus careful tuning of the sim.\n\nNo obvious circularity or load-bearing fitting issues show up. The work is for people in marine robotics and sim-to-real RL who want practical transfer examples. A reader focused on hardware validation will get value from the experiments.\n\nIt deserves serious referee time because the real-world results are there and the method is clearly described enough to evaluate.","headline":"The paper shows one RL policy with history-based latent inference transferring zero-shot to two real ASVs and beating non-adaptive baselines, but the simple analytical sim for randomization is the part that needs scrutiny.","tokens_in":2261,"tokens_out":405,"would_cite":false,"duration_ms":29758,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A single reinforcement learning policy trained in simulation generalizes to multiple real-world autonomous surface vehicles without fine-tuning.","keywords":["reinforcement learning","autonomous surface vehicles","trajectory tracking","cross-platform generalization","zero-shot deployment","adaptive control","teacher-student architecture"],"falsifier":"Deploy the same policy on a third autonomous surface vehicle whose dynamics lie outside the randomization ranges used in training and measure whether position tracking error remains comparable to the tuned controller.","tokens_in":2558,"feed_emoji":"🚤","tokens_out":692,"duration_ms":39011,"temperature":0.7,"pith_summary":"The paper sets out to establish that an adaptive reinforcement learning controller, trained once under randomized vessel dynamics in a basic simulation, can track trajectories on different physical boats without any retraining or platform-specific adjustments. It handles unknown dynamics by conditioning the policy on interaction history through a teacher-student module that infers a latent representation of each platform. This approach matters because most controllers require separate design and tuning for each vehicle due to differences in shape, mass, and water interaction. If the claim holds, one learned policy could serve varied platforms directly from simulation. Real-world tests on two platforms show the adaptive policy reduces position error by up to 58 percent compared with non-adaptive learning baselines and nearly matches a tuned controller built for one platform.","feed_headline":"One RL policy tracks trajectories on two different boats after simulation training","feed_subtitle":"Adaptive method with latent dynamics inference cuts position error up to 58% versus non-adaptive baselines on real platforms.","key_machinery":"The teacher-student architecture that infers a latent representation of unknown platform dynamics from interaction history and conditions the policy on that representation.","core_discovery":"The authors show that a policy trained in simulation with randomized vessel dynamics, using a teacher-student architecture to infer latent platform dynamics from interaction history, transfers zero-shot to two distinct real autonomous surface vehicles. The policy achieves position mean absolute error up to 58 percent lower than non-adaptive learning baselines while approaching the accuracy of a platform-specific tuned controller, even though the training model is a simple analytical dynamics approximation rather than a high-fidelity hydrodynamic simulator.","pith_inferences":["The same history-conditioning technique could support zero-shot transfer on other vehicle types whose dynamics vary across instances, such as different drone frames or ground robots.","Widening the randomization ranges or adding more platform parameters in simulation might further reduce the remaining gap to tuned controllers.","The results imply that partial-observability handling via latent inference is more critical for sim-to-real success than using high-fidelity physics models."],"forward_implications":["A single policy can be deployed across multiple autonomous surface vehicle platforms without per-platform retraining or tuning.","Training can use a basic analytical dynamics model instead of detailed hydrodynamic simulators while still achieving real-world generalization.","Adaptive policies conditioned on history can close much of the performance gap to hand-tuned platform-specific controllers.","Cross-platform transfer succeeds despite differences in actuation and hydrodynamic characteristics between the test vehicles."],"fun_headline_variants":["RL policy trained in sim transfers zero-shot to two real vessels","Adaptive RL infers latent dynamics for cross-platform boat control","Simulation with randomized dynamics enables zero-shot vessel transfer","Teacher-student RL achieves cross-boat tracking without fine-tuning"],"cache_read_input_tokens":64,"weakest_assumption_plain":"Randomizing vessel dynamics inside a simple analytical simulation model produces a policy that generalizes to the complex unmodeled hydrodynamic effects on real unseen platforms.","fun_headline_variants_meta":{"raw":{"variants":["RL policy trained in sim transfers zero-shot to two real vessels","Adaptive RL infers latent dynamics for cross-platform boat control","Simulation with randomized dynamics enables zero-shot vessel transfer","Teacher-student RL achieves cross-boat tracking without fine-tuning"]},"model":"grok-4.3","cost_usd":0.00421,"raw_usage":{"total_tokens":2100,"prompt_tokens":617,"num_sources_used":0,"completion_tokens":64,"cost_in_usd_ticks":42099500,"prompt_tokens_details":{"text_tokens":617,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1419,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":617,"tokens_out":64,"duration_ms":20299,"temperature":1.0,"reasoning_tokens":1419,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-03T11:51:48.329249+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Deploy the same policy on a third autonomous surface vehicle whose dynamics lie outside the randomization ranges used in training and measure whether position tracking error remains comparable to the tuned controller.","supporting_citations":[],"review_version":1}