{"id":"833e7101-1810-4463-b493-50b2659c1d86","arxiv_id":"2606.03441","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"PerchRL applies two-stage RL with randomized trajectories, temporal augmentation, and visibility-aware rewards to achieve vision-based perching on irregularly moving inclined platforms.","lead":"PerchRL is a two-stage reinforcement learning system that first pre-trains a quadrotor policy with full state information then fine-tunes it on camera images to perch on fast-moving tilted platforms. A general reader might examine it to see how RL can be made robust to intermittent vision and irregular motion in real drone tasks.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Generalization from randomized trajectories + temporal augmentation to real irregular motions lacks distributional validation","rationale":"The reader's weakest_assumption directly identifies the same unverified distributional match that underpins the adaptability and real-world claims. Full-text access does not alter this because the provided abstract and method summary contain no counter-evidence (distribution metrics or ablations). The concern is therefore load-bearing and internal to the argument rather than external consensus.","tokens_in":1647,"tokens_out":314,"duration_ms":13255,"concrete_test":"From the simulation setup section, extract the randomization ranges for platform velocity/acceleration/frequency; compute the same statistics on logged real-world platform trajectories (or video-derived estimates); if any first- or second-moment differs by >25% or the frequency content shows missing bands, retrain with matched randomization and re-measure perching success rate on the original real-world test set.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of adaptability without platform-specific tuning requires that randomized platform trajectories during pre-training plus temporal augmentation produce policies robust to the actual distribution of real-world irregular motions. The paper describes these techniques in the two-stage pipeline and hybrid vision fine-tuning but reports no quantitative comparison (e.g., power spectra, acceleration histograms, or KL divergence) between the training motion distribution and the real-world test trajectories, nor ablations isolating their effect on cross-platform success rates. Without this, real-world results could reflect limited test diversity or implicit tuning rather than the claimed generalization mechanism.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes PerchRL, a two-stage RL framework for vision-based agile perching of quadrotors on inclined platforms undergoing rapid and irregular motion. It consists of state-based pre-training on randomized trajectories with temporal augmentation to improve generalization, followed by vision-based fine-tuning using a hybrid framework with visibility-aware state augmentation and active perception rewards to handle intermittent visual loss. The central claim is that extensive simulation and real-world experiments demonstrate feasibility, stability, real-time performance, and adaptability across distinct quadrotor platforms without platform-specific tuning.","tokens_in":1755,"tokens_out":533,"duration_ms":19654,"significance":"If the experimental results hold with proper quantitative support, the work would contribute to agile aerial robotics by offering a generalizable approach to vision-based perching under challenging motion conditions, with potential applications in air-ground collaboration. The planned release of source code would support reproducibility in the field.","major_comments":[{"comment":"Abstract and §4 (Experiments): The central claim that 'extensive simulation and real-world experiments demonstrate the feasibility, stability, and real-time performance' is unsupported, as the manuscript provides no quantitative metrics, success rates, baseline comparisons, failure rates, or error bars to substantiate performance or generalization.","section":"Abstract, §4"},{"comment":"§3.2 (two-stage pipeline) and §3.3 (hybrid vision fine-tuning): The adaptability claim without platform-specific tuning rests on randomized trajectories plus temporal augmentation producing policies robust to real irregular motions, but no distributional validation (e.g., power spectra, acceleration histograms, or KL divergence between training and test motions) or ablations isolating their effect on cross-platform success is reported.","section":"§3.2, §3.3"},{"comment":"§4 (real-world deployment): The assertion of successful deployment across distinct quadrotor platforms validating adaptability lacks any reported quantitative cross-platform metrics or controls for test diversity, leaving open whether results reflect the claimed generalization or limited test conditions.","section":"§4"}],"minor_comments":[{"comment":"Notation for the temporal augmentation and active perception reward terms is introduced without explicit equations or parameter definitions, reducing clarity for readers attempting to reproduce the method.","section":"§3"},{"comment":"The abstract states source code will be released, but no link or repository is provided in the manuscript.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments, which highlight areas where the manuscript can be strengthened with additional quantitative support. We address each major comment below and will revise the manuscript to incorporate the suggested analyses and metrics.","responses":[{"response":"We agree that the abstract and experiments section would benefit from explicit quantitative metrics. In the revised manuscript, we will add tables reporting success rates, failure rates, baseline comparisons, and error bars from repeated trials to substantiate the claims of feasibility, stability, and real-time performance.","revision_made":"yes","referee_comment":"[Abstract, §4] Abstract and §4 (Experiments): The central claim that 'extensive simulation and real-world experiments demonstrate the feasibility, stability, and real-time performance' is unsupported, as the manuscript provides no quantitative metrics, success rates, baseline comparisons, failure rates, or error bars to substantiate performance or generalization."},{"response":"The referee is correct that distributional validation and targeted ablations are not currently reported. We will include comparisons of motion distributions (power spectra, acceleration histograms, KL divergence) between training and test sets, as well as ablation studies isolating the impact of randomized trajectories and temporal augmentation on cross-platform generalization.","revision_made":"yes","referee_comment":"[§3.2, §3.3] §3.2 (two-stage pipeline) and §3.3 (hybrid vision fine-tuning): The adaptability claim without platform-specific tuning rests on randomized trajectories plus temporal augmentation producing policies robust to real irregular motions, but no distributional validation (e.g., power spectra, acceleration histograms, or KL divergence between training and test motions) or ablations isolating their effect on cross-platform success is reported."},{"response":"We acknowledge the need for quantitative cross-platform metrics. The revision will report specific success rates, performance metrics, and controls for test diversity across the distinct quadrotor platforms to better support the adaptability claim.","revision_made":"yes","referee_comment":"[§4] §4 (real-world deployment): The assertion of successful deployment across distinct quadrotor platforms validating adaptability lacks any reported quantitative cross-platform metrics or controls for test diversity, leaving open whether results reflect the claimed generalization or limited test conditions."}],"tokens_in":1378,"tokens_out":491,"duration_ms":17339,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The paper introduces PerchRL as a two-stage RL method: state-based pre-training on randomized platform trajectories, followed by vision-based fine-tuning that adds visibility-aware state augmentation and active perception rewards to cope with intermittent views on inclined moving targets. Temporal augmentation is used to pick up motion patterns from history. This specific combination for agile perching under rapid irregular motion is presented as the contribution.\n\nIt addresses a concrete robotics need for air-ground collaboration and plans to release code, which is useful. The hybrid framework for handling vision loss during fine-tuning is a reasonable engineering step beyond plain RL.\n\nThe soft spot is the evaluation. The abstract asserts extensive simulation and real-world experiments plus cross-platform deployment, yet supplies no success rates, baselines, error bars, or timing numbers. The stress-test concern holds here: there is no reported check (spectra, histograms, or divergence) that the randomized training motions match the real irregular ones, so the generalization and no-tuning claims rest on unshown evidence. No load-bearing equations or fitted parameters appear that would make results circular.\n\nThis is for researchers working on vision-based drone control and RL for dynamic landing tasks. A reader could extract the pipeline structure and reward ideas even without the numbers. It deserves peer review because the problem is practical and the methods are described at a level that referees can assess, though the paper will need added quantitative results and ablations to stand up.","headline":"Two-stage RL pipeline with visibility tricks for perching on moving platforms, but abstract gives no metrics or comparisons to support the real-world claims.","tokens_in":2234,"tokens_out":360,"would_cite":false,"duration_ms":14004,"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 two-stage reinforcement learning method enables quadrotors to perch on inclined platforms moving rapidly and irregularly using only vision.","keywords":["reinforcement learning","vision-based perching","quadrotor control","agile flight","moving inclined platforms","robotics","autonomous landing"],"falsifier":"Real-world trials in which the platform follows an irregular trajectory outside the randomized training distribution produce repeated perching failures or loss of stability.","tokens_in":2572,"feed_emoji":"🚁","tokens_out":604,"duration_ms":15203,"temperature":0.7,"pith_summary":"Quadrotors struggle to perch on moving inclined surfaces because their cameras have a narrow field of view that frequently loses sight of the target. The paper presents PerchRL, which first pre-trains a policy with complete state data and then fine-tunes it on raw images, adding randomized platform paths during training and temporal augmentation of past observations to capture motion patterns. A visibility-aware augmentation step and rewards that encourage active looking help the policy recover when the platform disappears from view. Simulation and hardware tests show the resulting policies perch successfully in real time and transfer to different quadrotor bodies without retuning.","feed_headline":"Two-stage RL lets quadrotors perch on fast irregular slopes with vision","feed_subtitle":"Randomized training and temporal augmentation produce real-time stable policies that transfer across drone platforms.","key_machinery":"Two-stage RL pipeline that pre-trains on full state then fine-tunes on vision, with randomized platform trajectories, temporal augmentation of observations, visibility-aware state augmentation, and active perception rewards.","core_discovery":"PerchRL shows that a reinforcement learning policy trained in two stages—state-based pre-training followed by vision-based fine-tuning—can achieve stable agile perching on inclined platforms under rapid irregular motion when the training distribution includes randomized trajectories, temporal history augmentation, visibility-aware image augmentation, and active-perception rewards.","pith_inferences":["The same training recipe could be applied to perching on other surface orientations if the randomization range is expanded accordingly.","Temporal augmentation may allow the policy to anticipate short-term platform motion even without an explicit predictor.","Cross-platform success suggests the policy has captured platform-agnostic dynamics rather than hardware-specific parameters."],"forward_implications":["The learned policies run in real time on physical quadrotors.","The same policy transfers across distinct quadrotor platforms without retuning.","Successful perching occurs under both simulated and real rapid irregular platform motion.","The hybrid visibility and active-perception components maintain performance during intermittent visual loss."],"fun_headline_variants":["Two-stage RL perches quadrotors on irregular slopes using vision","Pretrain state then fine-tune vision for quadrotor incline perching","RL with randomized trajectories perches on rapid irregular platforms","Visibility augmentation in RL improves perching on moving inclines"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"Randomized platform trajectories during training plus temporal augmentation will produce policies that generalize to the distribution of real-world irregular motions without requiring platform-specific tuning.","fun_headline_variants_meta":{"raw":{"variants":["Two-stage RL perches quadrotors on irregular slopes using vision","Pretrain state then fine-tune vision for quadrotor incline perching","RL with randomized trajectories perches on rapid irregular platforms","Visibility augmentation in RL improves perching on moving inclines"]},"model":"grok-4.3","cost_usd":0.0104,"raw_usage":{"total_tokens":4570,"prompt_tokens":605,"num_sources_used":0,"completion_tokens":69,"cost_in_usd_ticks":103999500,"prompt_tokens_details":{"text_tokens":605,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3896,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":605,"tokens_out":69,"duration_ms":29658,"temperature":1.0,"reasoning_tokens":3896,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T09:24:37.657902+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Real-world trials in which the platform follows an irregular trajectory outside the randomized training distribution produce repeated perching failures or loss of stability.","supporting_citations":[],"review_version":1}