{"id":"4cfddc15-d537-4010-a64d-aa9800447ff9","arxiv_id":"2602.18803","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"LoTIS localizes a reference RGB trajectory in the robot's current view, predicting image-space coordinates, visibility, and distance to provide robot-agnostic guidance for navigation.","lead":"A new visual navigation system predicts where a reference path recorded on a phone would appear in a robot's current camera view, letting different robots follow that path without retraining or camera calibration. The system reports near-perfect success in simulated tests and in real trials on a drone and a quadruped, including backward traversal that other methods largely cannot do.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Real-world success claims rest on 3–6 trials per condition; without confidence intervals the 94–98% and 5× statements are not yet established.","rationale":"The reader's listed weakest assumption is the visual-overlap requirement. That is a real boundary condition, but the paper explicitly states it in Sec. III and in the Limitations, so it is a scoped limitation rather than a hidden threat to the central claim. The more load-bearing issue is statistical: the real-world experiments that anchor the abstract's '94–98% success' and '5× improvements' are based on very small trial counts, and the paper provides no uncertainty quantification. This is partially acknowledged in the reader's rationale ('small real-world trial counts'), so I mark agreement as partial. The verdict remains CONDITIONAL: the simulation results are strong enough to support the core method, but the real-world quantitative claims need either more trials or explicit confidence intervals before the abstract's unqualified wording is justified.","tokens_in":19956,"tokens_out":13819,"duration_ms":138787,"concrete_test":"Release per-trial logs for Tables II–III and compute exact 95% binomial confidence intervals for each condition. If lower bounds for individual real-world conditions fall below the abstract's '94–98%', revise the abstract to report intervals and scope. Alternatively, run a preregistered extension with ≥20 trials per condition (or a multi-seed simulation bootstrap) to obtain a tight CI for forward success; if the lower bound drops below about 90%, weaken the headline claim accordingly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central idea is coherent and the visual-overlap requirement is explicitly scoped in Sec. III, so it is not a hidden flaw. The load-bearing weakness is evidential: the abstract's quantitative real-world claims ('94–98% success', '5× improvements') rest on Table II and Table III, which use only 3–6 trials per condition and compare against a single baseline (FAINT). For n=6, a 6/6 success rate has a 95% Clopper-Pearson lower bound near 54%; for n=3, near 29%. The day→night backward condition is 2/3; crowded-env results are 5/6. The aggregate 35/36 is suggestive, but the reported point estimates are compatible with substantially lower true success rates. No confidence intervals, significance tests, per-trial logs, or seeds are reported. This does not invalidate the simulation evidence (Table I), but it undercuts the 'diverse sim and real' phrasing and leaves the real-world portion of the central claim less settled than the abstract implies.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes LoTIS, a learned perception module that, given a reference RGB trajectory and a current query image, predicts for each reference frame its image-space coordinates in the query view, a visibility logit, and a normalized distance. This representation is meant to be robot-agnostic and calibration-free, so that a simple yaw controller or an MPPI controller can use it for navigation on different embodiments. Training uses a cross-trajectory sampling strategy on real and simulated datasets, with labels generated from known poses and depth. The evaluation covers Gibson and HM3D simulation, comparing against ViNT, PlaceNav, NoMaD, and FAINT under matched/cross cameras and on/off-trajectory starts, as well as real-world trials on a Crazyflie quadrotor and a Spot quadruped using phone-recorded reference trajectories. The paper reports large simulation margins and high real-world success rates, and includes an explicit limitations section. The central claim is that decoupling perception from action in this way enables zero-shot embodiment transfer, backward traversal, and robustness to camera mismatch.","tokens_in":20216,"tokens_out":11640,"duration_ms":104314,"significance":"If the results hold, LoTIS is a meaningful step forward: the image-space trajectory-localization representation is a clean interface between learned perception and classical planning, the cross-trajectory training strategy is an elegant way to obtain viewpoint robustness without action labels, and the real-world deployment on both an aerial and a legged robot from phone-recorded trajectories is a valuable demonstration. The simulation study is substantial—200 runs per setup, multiple held-out scene datasets, and several strong baselines with official pretrained weights—and the paper is commendably candid about the visual-overlap requirement and other limitations. The availability of code and videos is a further strength. However, the real-world quantitative claims currently rest on very small trial counts, and the main ablation table is not clearly tied to the configuration reported in Table I. These issues need to be addressed before the claims as stated can be accepted.","major_comments":[{"comment":"The abstract's quantitative claims ('94–98% success rate across diverse sim and real environments', '5× improvements on backward traversal') rest in part on real-world results reported as point estimates from 3–6 trials per condition (Table II: 36 LoTIS runs, 18 per direction; Table III: 3–6 per robustness condition). No confidence intervals, significance tests, per-trial logs, or seeds are given. With n=6, a 6/6 success rate has a 95% Clopper-Pearson lower bound near 54%; with n=3, near 29%; the aggregate 35/36 has a lower bound near 85%. The reported point estimates are therefore compatible with substantially lower true success rates. The day→night backward cell (2/3) and the crowded-env backward cell (5/6) are particularly uncertain. This does not invalidate the simulation evidence (Table I, 200 runs per cell), but it undercuts the 'sim and real' phrasing and leaves the cross-embodime","section":"Abstract; Sec. VI-B, Tables II and III"},{"comment":"The values labeled 'LoTIS-F' in Table A1 are identical to the '+ Obstcl. Avoidance' row of Table I, not the base 'LoTIS (Ours)' row. For example, Gibson To End On-Trajectory Matched is 100.0 in Table A1 but 94.7 in Table I for LoTIS without obstacle avoidance. The appendix does not state that obstacle avoidance was enabled in the ablation. Since this table is the primary evidence for the contribution that full-trajectory joint processing is responsible for the gains, the configuration must be stated explicitly, and ideally the ablation should be reported for both controller variants so the reader can separate the effect of joint processing from the effect of obstacle avoidance.","section":"Appendix C, Table A1"}],"minor_comments":[{"comment":"The video list appears to swap table references: 'Real-World Navigation' is associated with Table III (which is the robustness table) and 'Robustness Studies' with Table II (which is the main real-world results table). Please correct the cross-references.","section":"Appendix B"},{"comment":"The caption has missing placeholders: 'starts at and ends at , with initial experiment positions shown as .' The symbols and legend are not rendered. Please fix the caption so the figure is self-contained.","section":"Fig. 4 caption"},{"comment":"The abstract says the system works 'without requiring camera calibration, poses, or robot-specific training.' The MPPI controller used for the quadrotor relies on UniDepthV2 to predict camera parameters and metric depth for collision avoidance and 3D grounding. Please clarify that the calibration-free claim applies to the LoTIS perception model itself, not to every downstream controller, or state the depth/camera assumption in the system-level claim.","section":"Abstract; Sec. IV-D and Appendix H-B"},{"comment":"The claim 'for the first time, using an RGB reference trajectory in the general setting...' is a strong novelty statement. Please either soften it or cite prior work that attempted this general setting and explain how the present contribution differs.","section":"Sec. I"},{"comment":"Minor typos and wording issues: 'where reference trajectory would appear' is missing an article; 'loosing visibility' in Appendix H-B should be 'losing visibility'; the phrase 'we achieve over 5x improvements' could be tightened to specify the comparison set. A careful proofread is recommended.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"This is a solid and original paper with a coherent central idea and strong simulation evidence. The main risk is overclaiming the real-world results given the small trial counts. I would not require a fully powered real-world study, but the abstract must be aligned with the evidence and the ablation configuration must be clarified. The Table A1/Table I discrepancy should be fixed before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a solid paper and should go to peer review, not be desk-rejected. The central idea is genuinely new: instead of mapping observations to actions or picking a single subgoal image, LoTIS predicts where every frame of a reference trajectory would appear in the robot's current view, along with visibility and normalized distance. That representation is embodiment-agnostic, it lets a downstream planner handle things like backward traversal, and the cross-trajectory training strategy is a sensible way to get the viewpoint and camera robustness they claim.\n\nThe simulation evaluation is the strongest part. Gibson and HM3D, 100 reference trajectories each, matched and cross-camera conditions, on- and off-trajectory starts, forward, backward, and any-point tasks, and they beat ViNT, NoMaD, PlaceNav, and FAINT by large margins (20–50 percentage points in most rows). The real-world transfer to a quadrotor and a Boston Dynamics Spot using phone-recorded trajectories is a genuine demonstration, not just a video of cherry-picked runs. They also explicitly scope the method in the Limitations: it needs visual overlap with some portion of the reference trajectory, and it degrades with large height mismatches. That is honest and load-bearing; the method won't work if the robot can't see any part of the path.\n\nWhere the paper is soft is the real-world statistics, and the stress-test note is right about that. Table II shows 35/36 for LoTIS against 9/36 for FAINT, but those are per-condition n of 3 or 6. A 6/6 success rate has a 95% Clopper-Pearson lower bound around 54%; 3/3 is around 29%. No confidence intervals, no per-trial logs, no seeds. The abstract's \"94–98% success across diverse sim and real\" and \"5x improvements\" overstate what a fair reading of the real data supports. The sim results, with n=200 per condition and huge margins, are the load-bearing evidence; the real trials corroborate but do not by themselves establish those numbers. There is also a small arithmetic inconsistency: the text says FAINT succeeds in 27.8% of trials, but summing Table II gives 9/36 = 25%. That should be fixed.\n\nOne more thing to check in review: the baselines don't get the added obstacle-avoidance module LoTIS is paired with in some rows, though they do report LoTIS without it, so it's not a hidden confound.\n\nWho gets value: anyone working on visual navigation, visual place recognition, or learning from demonstration. It deserves a careful referee. I would send it out and ask for error bars or confidence intervals on the real-world claims, the corrected arithmetic, and a clean code/data release with a clear commit.","headline":"LoTIS's core idea—predicting the whole reference trajectory in image-space—is a real advance, and the sim evidence is strong; the real-world numbers need statistical humility.","tokens_in":20672,"tokens_out":3436,"would_cite":true,"duration_ms":31753,"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":"LoTIS shows that visual navigation can be reduced to localizing a recorded reference trajectory in the robot's current camera view, and that this image-space guidance alone drives different robots along a route, forward or backward, without","keywords":["visual navigation","reference trajectory","image-space localization","robot-agnostic guidance","cross-trajectory training","transformer","backward traversal","camera mismatch"],"falsifier":"Start a robot in a view that shares no visible content with any frame of the reference trajectory—for instance, on the far side of a sharp corner or behind a closed door—and check whether the model reports zero visible trajectory points and the robot stalls; the paper's overlap assumption predicts exactly this failure, so a successful traversal from such a start would show the assumption is not load-bearing.","tokens_in":19906,"feed_emoji":"🧭","tokens_out":10420,"duration_ms":81440,"temperature":0.7,"pith_summary":"This paper tries to establish that visual navigation should be split into perception and control: rather than learning to turn images into actions, a model learns to predict where the poses of a recorded reference route would appear in the robot's current camera view—as 2D points, visibility flags, and normalized distances. The claim is that this image-space representation is enough for an off-the-shelf local planner to drive the robot along the route, and that because the output lives in the image, the same prediction works for different robot platforms, cameras, and even backward traversal. The paper backs this with a cross-trajectory training scheme that deliberately mismatches cameras and viewpoints, and with simulation and real-world experiments showing 94–98% forward success and large gains over action-learning baselines, plus a demonstration that one phone-recorded video guides both a quadrotor and a quadruped. If the claim is right, recording a route once with any camera makes it navigable by any robot, to any point on the route, in either direction, without calibration.","feed_headline":"One route video steers any robot, forward or backward","feed_subtitle":"The same phone-recorded route guides quadrotors, quadrupeds, and mismatched cameras without retraining.","key_machinery":"The load-bearing object is the image-space trajectory representation: for each reference frame i, the model predicts a 2D point p_i where that frame's camera pose would appear in the current view, a visibility logit v_i, and a normalized distance d_i. The architecture is an asymmetric transformer: a trajectory encoder processes all reference frames jointly with alternating global and per-frame attention, a lightweight query encoder handles the current view, a fusion module lets each trajectory token attend to the query view to establish correspondences, and a recurrent head progressively refines the predictions over several iterations. Training uses cross-trajectory sampling—reference and qu","core_discovery":"The paper's central discovery is that navigation guidance can be reduced to a single perceptual prediction: for every frame in a reference RGB trajectory, estimate where that frame's viewpoint sits in the robot's current image, whether it is visible, and how far away it is. A transformer architecture processes the full trajectory jointly (rather than choosing one subgoal image), fuses it with the current view, and iteratively refines these predictions. Because the output is defined in the robot's own image frame, any downstream planner—from a yaw-only controller to a model-predictive one—can consume it directly. Trained with a cross-trajectory strategy in which reference and query views come","pith_inferences":["The decoupling principle points toward a modular navigation stack where perception and planning improve independently; one testable extension is training the model to emit a heading or 'pointing' signal when the trajectory is out of view, which the paper lists as future work.","The visual-overlap dependency suggests the method is inherently a local navigation oracle; scaling to long routes will likely require memory or recovery mechanisms, which the paper acknowledges via trajectory chunking.","Cross-trajectory sampling could be exported to other vision-for-robot tasks that currently need action-labeled demonstrations, potentially shrinking the data burden for embodiment-agnostic perception.","One might test whether the model is implicitly learning the 3D shape of the trajectory: if it does, its predictions could be used to reconstruct path geometry or for localization in repetitive environments where appearance matching is unreliable."],"forward_implications":["A single phone-recorded route video can serve as a navigation instruction for any camera-equipped robot, without recalibration or retraining.","Navigation to any point along a route—including backward traversal—becomes possible from starting positions that are off the route, provided some part of the route is visible.","Mismatched cameras (field of view, aspect ratio, mounting height) between the recording and the robot no longer cause navigation to fail, within the ranges tested in the paper.","Because the output is image-space guidance, the same perception model can be paired with different local planners and action spaces, including aerial and legged platforms.","The full-trajectory joint processing avoids the subgoal-selection failures of prior methods, so localization accuracy and success rate degrade much more slowly with distance from the route."],"fun_headline_variants":["One route video steers any robot, even backward","Single video guides any robot without retraining","Robot-agnostic: one video, any camera, any bot","From phone video to robot navigation, zero calibration","One video to steer any robot, forward or reverse"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The method assumes that at every decision step some part of the reference trajectory is visible in the robot's current camera view; when that overlap disappears (sharp backward corners, extreme height mismatch, or occlusion), the model cannot produce guidance and navigation degrades—a limitation the paper explicitly concedes.","fun_headline_variants_meta":{"raw":{"variants":["One route video steers any robot, even backward","Single video guides any robot without retraining","Robot-agnostic: one video, any camera, any bot","From phone video to robot navigation, zero calibration","One video to steer any robot, forward or reverse"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000222,"raw_usage":{"total_tokens":1292,"prompt_tokens":748,"completion_tokens":544,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":492,"completion_tokens_details":{"reasoning_tokens":468}},"tokens_in":492,"tokens_out":544,"duration_ms":5020,"temperature":1.0,"reasoning_tokens":468,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T21:49:51.239724+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Start a robot in a view that shares no visible content with any frame of the reference trajectory—for instance, on the far side of a sharp corner or behind a closed door—and check whether the model reports zero visible trajectory points and the robot stalls; the paper's overlap assumption predicts exactly this failure, so a successful traversal from such a start would show the assumption is not load-bearing.","supporting_citations":[],"review_version":1}