{"id":"0dbbe1e7-91f9-49c1-a57d-858fa295e4a1","arxiv_id":"2611.02474","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A world-action model that jointly predicts video, actions, and a hand-object 4D motion field, and steers its attention to interactions and grasp/release frames, tops three simulation benchmarks and a five-task real-robot test, most clearly under visual shifts.","lead":"Gnomon4D is a robot controller that predicts future video, its own actions, and a geometric motion field of the hand and objects, then uses that field to focus its decisions on moments when the hand touches an object. It reports top results on three robot simulation benchmarks and on a five-task real robot, with the largest edge under changed lighting, textures, and distractors.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Benchmark wins do not isolate the GeoDelta-readout mechanism from privileged test-time inputs (depth + mask; joint-state renders on RoboDojo); the RoboDojo 2× margin and VLABench +7.4 texture gain could be input-driven, and the reader's real-robot DA3 worry misreads that leg — it uses measured…","rationale":"I read the paper's central claim as causal: GeoDelta jointly predicted with video and actions, read where and when interactions occur, and pretrained in a shared human-robot motion space, is the recipe behind the SOTA numbers. The three legs of evidence are internally consistent — every headline rate reproduces from the supplement's success counts (165/200 ID, 156/200 OOD; Table 10), the ablation increments are arithmetically coherent, and the real-robot protocol (shared initial configurations, alternating trials, per-task and per-shift counts) is among the more careful of its kind. The paper also deserves credit for disclosing the label-estimation pipeline in its Limitations and for reporting the condition inputs factually in Secs. 3.4 and 4.1. The load-bearing unmet condition is attribution: the Action expert attends directly to current depth+mask condition tokens, so the benchmark comparisons conflate 'reading predicted GeoDelta' with 'being handed clean geometry at each replan.' The two results the paper most leans on — the RoboDojo 2× margin and the VLABench unseen-texture robustness — are exactly the legs with no geometry-conditioned baseline, and on RoboDojo the conditions are rendered from live joint states. The reader's chosen weakest assumption (DA3 on the real robot) does not survive contact with Sec. 4.1 and F.2, which specify measured RGB-D depth for that leg; the DA3 rescaling concern applies to RoboDojo object GeoDelta (Eq. 8) and Robot-30H mid-training. Because Table 4b gives the readout +2.6 over generic queries with inputs held fixed, the mechanism has genuine internal support; the right conclusion is not rejection but a demand for an isolation experiment, which is precisely the conditionality the reader's verdict already encodes. I therefore keep the verdict at CONDITIONAL (UNCHANGED) while shifting the emphasized risk from label estimation to input isolation.","tokens_in":34903,"tokens_out":26003,"duration_ms":191987,"concrete_test":"Run the Table-4-style cross on RoboDojo: train the 2×2 {joint-state depth+mask condition inputs on/off} × {generic-query dense readout / InSAttn+TransAgg} with pretraining fixed, and fine-tune OpenWAM-Alpha on the same depth+mask conditions. If the readout's margin over generic queries collapses when the conditions are removed, or if OpenWAM-Alpha with those conditions approaches the 23.86% SR, then the RoboDojo margin and its 'precision via depth' attribution reflect test-time geometric conditions, not the GeoDelta mechanism.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The least-secure condition for the paper's central claim — that reading a jointly predicted GeoDelta at interaction regions and transition frames drives the benchmark wins — is isolation of that mechanism from the model's test-time inputs. In Sec. 3.4, the Action expert cross-attends directly to the clean condition tokens V_obs and G_cond (current depth + foreground mask); it does not need the predicted motion field to see geometry. The headline comparisons against RGB baselines therefore bundle three differences: (i) privileged condition inputs, (ii) the predicted GeoDelta plus focused readout, and (iii) pretraining. On RoboDojo, the conditions are kinematically exact robot renders recomputed at test time from live joint states (Sec. 4.1, E.1) — the strongest possible geometric prior — and the headline is a 2× success-rate gain (23.86 vs 11.92) whose Precision story (28.24 vs 20.42) is credited to GeoDelta encoding 'exact positions in both the image plane and depth.' On VLABench, the +7.4 unseen-texture margin is credited to 'GeoDelta encodes motion rather than appearance,' but the simulator-rendered depth+mask conditions are equally appearance-invariant, so that robustness admits a direct input-level explanation. The reader's weakest assumption is partly off target: the real-robot leg uses measured RGB-D depth, not DA3 (Sec. 4.1, F.2); DA3 enters only RoboDojo object GeoDelta (Eq. 8, one fitted scale per episode) and Robot-30H mid-training, and the paper discloses that estimation risk in its Limitations. The RoboTwin leg is partially controlled — X-WAM is RGB-D yet trails by 4.11 — and Table 4b shows InSAttn+TransAgg adds +2.6 over generic queries with conditions fixed, so the mechanism is not vacuous. What is missing is any experiment crossing condition-input presence with readout type; without it, the share of the flagship RoboDojo and VLABench margins creditable to reading predicted GeoDelta, rather than to handed-in geometry, is unquantified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"Gnomon4D is a mixture-of-transformers world-action model (WAM) that jointly denoises future video, a 32-step action chunk, and GeoDelta, a video-aligned 4D field (Δu, Δv, Δd) encoding the image-plane displacement and depth change of embodiment and object surface points between sampled frames. A relative-motion field derived from predicted GeoDelta features feeds two readout mechanisms: Interaction-Salient Attention (InSAttn) biases the action expert's spatial attention toward regions where the embodiment and objects move relative to each other, and Transition-Anchored Aggregation (TransAgg) pools the readout over frames where that relation changes (grasp and release). Because GeoDelta lives on the video pixel grid rather than in embodiment-specific control space, the authors use it as a human-robot quasi-action to pretrain the Video and GeoDelta experts on EgoManip-5000H (curated from a proprietary library) and to mid-train the full model on Robot-30H before post-training on each benchmark. Experiments report the best average success rate on RoboTwin 2.0 (94.33%), VLABench (62.5%), and RoboDojo (23.86%), and on a real UR5e platform 82.5% in-distribution and 78.0% OOD success over five tasks (165/200 and 156/200), versus 75.5%/62.0% for the best baseline. Ablations on 10 RoboTwin tasks attribute 3.4 points to full GeoDelta over RGB under dense readout, 2.6 points to the guided readout over generic queries, and 4.8 points to the two pretraining stages.","tokens_in":35287,"tokens_out":15030,"duration_ms":104652,"significance":"If the results hold, the paper's recipe — jointly predicting a 4D embodiment-object motion field and gating the action readout on interaction-salient regions and transition frames, then pretraining on human video in that shared motion space — is a genuinely new organizational principle for WAM design, and the real-robot OOD margin (78.0% vs 62.0% for the best baseline, Table 11) is a practically substantial result. The paper is unusually careful on several axes that often sink this literature: the real-robot protocol uses shared initial configurations, alternating trials, and complete success counts (Tables 10–11); GeoDelta is predicted rather than oracle-fed at inference; the Table 4 ablations are diagnostic (InSAttn adds 1.7 points over generic queries under identical conditions, a contrast the mechanism's construction does not guarantee); and the attention probe reports bootstrap intervals. I do not share the reader's concern about the real-robot depth channel, since Sec. F.2 documents measured sensor depth; the estimation concern is specific to RoboDojo object GeoDelta and to Robot-30H mid-training.","major_comments":[{"comment":"The benchmark margins credited to GeoDelta are not isolated from the model's test-time condition inputs. In Sec. 3.4 the Action expert cross-attends directly to the clean tokens V_obs and G_cond (current depth and foreground mask), so a geometric input channel is available even if the predicted GeoDelta were never read. On RoboDojo (E.1), G_cond is a robot-only depth render recomputed at test time from live joint states and camera parameters — the strongest possible geometric prior — yet Sec. 4.2 credits the 23.86 vs 11.92 SR margin and the 28.24 vs 20.42 Precision gain to GeoDelta encoding 'exact positions in both the image plane and depth.' Similarly, the VLABench claim that the margin over OpenWAM-Alpha grows from 0.6 to 7.4 points on unseen textures 'as GeoDelta encodes motion rather than appearance' is not established by a comparison against an RGB-only baseline that lacks the appearance-invariant depth+mask conditions. Table 4 is the right kind of evidence — it holds the conditions fixed and still shows 2.6 points for InSAttn+TransAgg and 3.4 points for full GeoDelta over RGB — but it is restricted to 10 RoboTwin 2.0 tasks. Please add an input-control experiment on VLABench and RoboDojo (for example, ablating G_cond, or providing the baselines with the same rendered depth/mask conditions), or reword the Sec. 4.2 attributions so the margins are described as joint effects of the condition inputs and the GeoDelta readout.","section":"Sec. 3.4, 4.2, E.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The RoboDojo Precision claim rests on a depth-change channel whose supervision is estimated, and the magnitude of the resulting risk is unquantified. Object GeoDelta on RoboDojo uses DA3 monocular depth rescaled by a single per-episode factor s fitted against rendered robot depth (Eq. 8); if s is biased for objects at depths different from the arm, or if the DA3/RAFT correspondences drift, the Δd channel — precisely the channel invoked to explain the 28.24% Precision result — is trained on a biased signal. The Limitations paragraph acknowledges that estimation errors 'mainly perturb Δd,' and the real-robot leg is indeed not the main exposure since it uses measured sensor depth (Sec. F.2, Table 7). The RoboDojo leg is where the risk is largest, and no sensitivity analysis is provided. A concrete fix would be to validate s on episodes with known object geometry, or to retrain the RoboDojo policy with Δd zeroed or corrupted and show that the Precision and overall gains depend on the estimated depth change rather than on the image-plane motion and the rendered conditions. Without such an analysis, the RoboDojo results remain compatible with a label-artifact explanation.","section":"Eq. (8), Sec. E.1, Sec. 4.2 (RoboDojo)","authors_acknowledge":"Limitations and future work.Without simulator state, robot GeoDelta is estimated from measured or monocular depth, masks, and optical flow, whose errors on thin, reflec- tive, or transparent surfaces mainly perturb ∆d.","authors_acknowledge_p":0.71}],"minor_comments":[{"comment":"The depth-change scale ω_d is only described as set from the depth-change statistics of the training data; please report the value per domain (the real-robot section uses ω_d = 0.10 m in F.2) and add a brief sensitivity check for ω_d and the soft thresholds (0.5 px and 2 mm), since the Δd channel is the paper's core differentiator.","section":"Sec. 3.3, Eq. (2)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"With 40 trials per task, per-task margins of two to five successes are within binomial sampling noise, and the aggregate ID margin (165 vs 151 of 200) is roughly 1.7 standard errors, whereas the OOD margin (156 vs 122) is highly significant; please report binomial confidence intervals and either soften the 'ranks first on every task' claim or anchor the strong claim to the OOD aggregate.","section":"Sec. 4.3, Tables 10–11","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The main text should state up front that on RoboDojo the model's test-time conditions are robot-only renders computed from live joint states, since the current wording leaves this input-privilege detail in the supplementary only.","section":"Sec. 4.2, E.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The attention probe compares Gnomon4D against baselines but does not include a Gnomon4D variant without InSAttn and TransAgg, so the probe difference could in principle reflect pretraining or the GeoDelta supervision rather than the readout; adding such a variant would make the probe diagnostic for the proposed mechanism.","section":"Sec. A (Supplementary)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"EgoManip-5000H is mined from an industrial-scale proprietary video library; please state whether the pretraining corpus, the Robot-30H processing, or the trained checkpoints will be released, since this directly determines the reproducibility of the pretraining-transfer claim.","section":"Sec. 3.6, D.1","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Two things to know before you read Gnomon4D. The novel piece is real: GeoDelta, a jointly generated 4D motion field (Δu, Δv, Δd) that puts embodiment and object points on one pixel grid so relative motion marks an interaction, and the InSAttn/TransAgg readout that reads it at interaction regions and transition frames. I checked the paper's own taxonomy and the cited work, and no baseline gives the action online access to joint embodiment–object 3D motion. The real-robot evaluation is also better than most of this subfield: shared initial configurations, alternating trials, and full per-trial counts (165/200 ID and 156/200 OOD, vs 124/200 for the best baseline). That 16-point OOD gap is the strongest result in the paper.\n\nCredit where it is due: the ablations are genuinely diagnostic. With conditions fixed, full GeoDelta beats depth by 2.5 points and flow by 0.9, object motion drives the gain, and InSAttn + TransAgg adds 2.6 over generic queries. The readout mechanism is not vacuous, and the probe's baseline side — released WAMs giving 1.8–2.3× attention to the non-interaction embodiment — is a legitimate motivation, even if Gnomon4D's own side partly restates its design.\n\nThe soft spots, in order of size. First, the headline margins on RoboDojo and VLABench do not isolate the mechanism from the model's inputs. The Action expert cross-attends directly to clean depth + mask conditions, and on RoboDojo those conditions are kinematically exact robot renders recomputed from live joint states at test time. The 2× RoboDojo margin and the +7.4 VLABench texture gain both admit input-level explanations: appearance-invariant handed-in geometry can produce texture robustness without any motion field. The paper never crosses condition-input presence with readout type, so the mechanism's share of its biggest numbers is unquantified. To be fair, the handed-in geometry on RoboDojo and VLABench covers the robot, not the objects, and the RoboTwin leg partially controls (X-WAM is RGB-D and trails by 4.11), but a referee should demand the missing cross experiment.\n\nSecond, the reader's DA3 worry is half off target. The real robot uses measured RealSense RGB-D depth, not DA3 monocular depth; DA3 enters only RoboDojo object GeoDelta with one fitted scale per episode (Eq. 8) and the Robot-30H mid-training. That estimation risk is real, and the paper discloses it in Limitations, but it should not be loaded onto the flagship real-robot numbers.\n\nThird, minor: the RoboTwin margins are under a point with no error bars, and the full pipeline is unauditable — no code or checkpoints, and EgoManip-5000H is proprietary, so the pretraining contribution in Table 4c cannot be independently verified.\n\nThis deserves a serious referee — anyone working on world-action models or geometry-conditioned policies should engage with it. The right outcome is major revision asking for an input-matched control; if that lands, this is a solid contribution, and even as it stands the GeoDelta representation and the real-robot protocol are worth citing.","headline":"A well-built WAM paper whose GeoDelta representation and real-robot protocol are the real news, but its biggest benchmark margins are confounded by privileged test-time inputs, so the mechanism's true share is unquantified.","tokens_in":36170,"tokens_out":9696,"duration_ms":58298},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"The paper claims precise robot control comes from where and when a policy reads geometric motion: read a joint robot-object 4D motion field at contact regions and grasp-release frames, and manipulation succeeds even when the scene looks…","keywords":["world-action model","GeoDelta","4D motion field","robot manipulation","grounded action readout","human-video pretraining","mixture-of-transformers","visual robustness"],"falsifier":"In simulation, retrain the RoboDojo policy with the $\\Delta d$ channel of the object GeoDelta labels zeroed while keeping $(\\Delta u,\\Delta v)$, and check whether the Precision-dimension success rate of 28.24% and the overall margin over OpenWAM-Alpha (23.86% vs. 11.92%) survive; on the real robot, redo Charger Insertion and Color Sorting with GeoDelta labels computed from calibrated multi-view stereo or motion capture instead of DA3 depth plus RAFT flow, and check whether the 78.0% out-of-distribution average and the 16-point margin over the best baseline persist.","tokens_in":38630,"feed_emoji":"🤖","tokens_out":3060,"duration_ms":80644,"temperature":0.7,"pith_summary":"Gnomon4D claims that robot world-action models fail at precise manipulation not because they lack geometry, but because their action module reads the predicted future at the wrong place and time. The paper's answer is GeoDelta, a video-aligned 4D motion field that records, for every foreground pixel of the robot and the manipulated object, the image-plane displacement and depth change of the same physical point. Because embodiment and objects share one field, their relative motion marks where an interaction happens and when it transitions at grasp and release, and two attention mechanisms use that cue to focus action readout. The paper reports state-of-the-art success rates on RoboTwin 2.0, VLABench, and RoboDojo, and a real-robot edge that grows under visual shifts, from 82.5% in-distribution to 78.0% out-of-distribution while the best baseline falls from 75.5% to 62.0%. A reader would care because the paper's claim identifies a mechanism, motion-guided readout, rather than more data or bigger models, as the source of control accuracy and visual robustness.","feed_headline":"Reading 4D motion at contact beats RGB-only robot world models","feed_subtitle":"Gnomon4D predicts a shared robot-object motion field and reads it at grasp and release, keeping success under visual shifts.","key_machinery":"The load-bearing object is GeoDelta: a video-aligned 4D geometric motion field that stores at each foreground pixel the displacement $(\\Delta u,\\Delta v)$ and depth change $\\Delta d$ of the same physical point between adjacent sampled frames, measured in the source camera so camera motion cancels, and encoded with signed $\\mu$-law companding ($\\mu=31$) into three RGB channels of the frozen video VAE. From GeoDelta features, lightweight heads predict per-token motion $\\delta$ and embodiment/object role probabilities, and the relative-motion field $c = \\|\\delta - \\bar\\delta_e\\|/(\\|\\delta\\| + \\|\\bar\\delta_e\\|)$ measures how far a token departs from its local embodiment neighborhood, so it is near 1 at interactions and near 0 when things move together, with a frame-to-frame change that spikes at grasp and release. Interaction-Salient Attention (InSAttn) converts $c$ into a task-conditioned saliency bias for action tokens, Transition-Anchored Aggregation (TransAgg) scores frames by saliency-weighted change of $c$ and pools around transitions, and the resulting interaction tokens enter the Action expert residually. A shared denoising clock, with $\\sigma_V(u)=1-u$ and $\\sigma_G(u)=\\sigma_A(u)=\\max(1-2u,0)$, finishes GeoDelta and actions before video refinement, so the action reads the predicted motion as it forms.","core_discovery":"The paper's central claim is that the usefulness of a geometric future for robot control depends on how the action reads it, not only on what the future encodes. Probing three released world-action models on RoboTwin 2.0 shows their action attention allocates 1.82-2.30 times uniform attention to the non-interacting rest of the embodiment versus 1.18-1.52 times to the interaction region; Gnomon4D reverses this to 0.74 times and 2.42 times by generating GeoDelta, the per-pixel $(\\Delta u,\\Delta v,\\Delta d)$ motion of embodiment and object points on one grid, and gating the action readout with the predicted embodiment-object relative-motion field. The paper argues, and reports evidence, that this mechanism together with GeoDelta as an embodiment-agnostic quasi-action that enables pretraining on 5,000+ hours of human egocentric video yields the best average result on all three simulation benchmarks, with 94.33% on RoboTwin 2.0, 62.5% on VLABench, and 23.86% on RoboDojo, and a real-robot 82.5% in-distribution and 78.0% out-of-distribution average. Ablations attribute the gain mainly to the object-motion component of GeoDelta and to the guided readout over generic queries, supporting the paper's thesis that a predicted future serves control through both what it encodes and how the action reads it.","pith_inferences":["Editorial inference: the where-and-when readout could transfer beyond tabletop manipulation to contact-rich assembly, locomotion with ground contact, or human-robot handover, wherever a foreground relative-motion field marks an interaction; the paper does not test these settings.","Editorial inference: the paper's ablations replace GeoDelta wholesale with depth or 2D flow but never decompose the $\\Delta d$ channel within GeoDelta itself, so a sharper experiment that zeroes or corrupts $\\Delta d$ while keeping $(\\Delta u,\\Delta v)$ would separate the contribution of line-of-sight motion from image-plane motion.","Editorial inference: because InSAttn and TransAgg act as a residual readout on predicted motion, they could plausibly be stacked with end-effector-distance biases or interaction-aware training objectives without conflict, a combination the paper does not explore.","Editorial inference: the real-robot and RoboDojo legs of the results rest on estimated labels, so a skeptic would want the real-robot protocol repeated with ground-truth depth or motion capture; the simulator legs already use exact rendered GeoDelta and isolate the mechanism in simulation."],"forward_implications":["If the central claim holds, any world-action model that keeps a video prior can gain precision and visual robustness by generating an embodiment-object 4D motion field at inference and reading it at interaction-salient regions and transition frames.","Human egocentric video becomes a usable pretraining source for robot policies without hand-pose retargeting, because GeoDelta is a shared motion space for hands and grippers; the paper reports a 2.1-point gain from EgoManip-5000H pretraining and a further 2.7 points from Robot-30H mid-training.","Robustness to unseen appearance follows from the mechanism: GeoDelta encodes motion and depth change rather than texture, and the paper's largest real-robot out-of-distribution margins, 18 points under unseen object appearance and added distractors, occur exactly where texture-invariant motion is unchanged.","Precision manipulation gains the most: on RoboDojo's Precision dimension the paper reports 28.24% success versus at most 20.42% for the baselines, consistent with GeoDelta encoding image-plane and depth motion jointly.","Action attention can be deliberately steered rather than left to emerge: probes show released WAMs underweight the interaction region, and Gnomon4D reverses that allocation, implying that attention probing is a usable diagnostic for where world-action models look."],"supporting_citations":[{"why":"Supplies the frozen video VAE and the Wan2.2-TI2V-5B initialization of the Video expert, the video prior that GeoDelta reuses for encoding.","marker":"[65]"},{"why":"Mixture-of-Transformers is the modality-expert architecture Gnomon4D extends with its GeoDelta expert and the shared joint attention.","marker":"[35]"},{"why":"DA3 monocular depth is used to estimate object GeoDelta on RoboDojo and robot GeoDelta on the real robot, where ground-truth depth is unavailable.","marker":"[36]"},{"why":"RAFT optical flow supplies the cross-frame correspondences needed to compute estimated GeoDelta labels and its forward-backward validation.","marker":"[62]"},{"why":"SAM 3.1 provides the embodiment and object masks that define GeoDelta foreground pixels and role supervision where simulator labels are absent.","marker":"[7]"},{"why":"RoboTwin 2.0 is the main bimanual benchmark, replayed from seeds to produce exact GeoDelta supervision and used for the attention probes.","marker":"[11]"},{"why":"RoboDojo is the benchmark whose object motion is estimated with DA3 depth rescaled by one fitted factor per episode, the site of the paper's depth-scale assumption.","marker":"[10]"},{"why":"VLABench is the benchmark behind the unseen-texture robustness claim, where the margin over OpenWAM-Alpha grows from 0.6 to 7.4 points.","marker":"[87]"},{"why":"OpenWAM-Alpha is the strongest RGB-future WAM baseline on all three benchmarks and one of the models probed for action attention.","marker":"[70]"},{"why":"FlowWAM is the RGB-plus-2D-flow baseline on the real robot whose in-distribution lead (75.5%) collapses under visual shifts to 62.0%.","marker":"[12]"}],"fun_headline_variants":["Where and when to read 4D motion for robot action","World-action model reads 4D motion at grasp and release","Focused 4D motion readout outperforms RGB-only world models","Gnomon4D reads shared motion where robot meets object","4D motion field read at contact beats RGB robot baselines"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The load-bearing premise is that estimated labels stand in for true 4D motion: on the real robot and in RoboDojo, GeoDelta is computed from monocular depth, segmentation masks, and optical flow rather than measured ground truth, so if those estimates are biased, the depth-change signal that drives the reported precision and out-of-distribution gains teaches the policy a wrong motion, an estimation issue the paper acknowledges in its Limitations section.","fun_headline_variants_meta":{"raw":{"variants":["Where and when to read 4D motion for robot action","World-action model reads 4D motion at grasp and release","Focused 4D motion readout outperforms RGB-only world models","Gnomon4D reads shared motion where robot meets object","4D motion field read at contact beats RGB robot baselines"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000185,"raw_usage":{"cost":0.0001852214,"is_byok":false,"cost_details":{"upstream_inference_cost":0.0001852214,"upstream_inference_prompt_cost":4.12594e-05,"upstream_inference_completions_cost":0.000143962},"total_tokens":2050,"prompt_tokens":1259,"completion_tokens":791,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":1007,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":717}},"tokens_in":1259,"tokens_out":791,"cost_source":"openrouter_usage","duration_ms":8702,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":717,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-05T05:17:22.339928+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"In simulation, retrain the RoboDojo policy with the $\\Delta d$ channel of the object GeoDelta labels zeroed while keeping $(\\Delta u,\\Delta v)$, and check whether the Precision-dimension success rate of 28.24% and the overall margin over OpenWAM-Alpha (23.86% vs. 11.92%) survive; on the real robot, redo Charger Insertion and Color Sorting with GeoDelta labels computed from calibrated multi-view stereo or motion capture instead of DA3 depth plus RAFT flow, and check whether the 78.0% out-of-distribution average and the 16-point margin over the best baseline persist.","supporting_citations":[{"cited_title":"Track4Action: Distilling World-Centric 3D Tracker into Vision-Language-Action Policies","cited_arxiv_id":"2608.03727","evidence_quote":"Supplies the frozen video VAE and the Wan2.2-TI2V-5B initialization of the Video expert, the video prior that GeoDelta reuses for encoding."},{"cited_title":"Breaking the Vision-Action Shortcut: Latent Interface Training for Generalizable Robotics Foundation Models","cited_arxiv_id":"2609.12641","evidence_quote":"Mixture-of-Transformers is the modality-expert architecture Gnomon4D extends with its GeoDelta expert and the shared joint attention."},{"cited_title":"Chen, Zhenyu Li, Yang Zhao, Sida Peng, Hengkai Guo, Xiaowei Zhou, Guang Shi, Jiashi Feng, and Bingyi Kang","cited_arxiv_id":null,"evidence_quote":"DA3 monocular depth is used to estimate object GeoDelta on RoboDojo and robot GeoDelta on the real robot, where ground-truth depth is unavailable."},{"cited_title":"PointAction: 3D Points as Universal Action Representations for Robot Control","cited_arxiv_id":"2606.03943","evidence_quote":"RAFT optical flow supplies the cross-frame correspondences needed to compute estimated GeoDelta labels and its forward-backward validation."},{"cited_title":"SAM 3: Segment anything with concepts","cited_arxiv_id":null,"evidence_quote":"SAM 3.1 provides the embodiment and object masks that define GeoDelta foreground pixels and role supervision where simulator labels are absent."},{"cited_title":"FlowWAM: Optical Flow as a Unified Action Representation for World Action Models","cited_arxiv_id":"2607.13017","evidence_quote":"RoboTwin 2.0 is the main bimanual benchmark, replayed from seeds to produce exact GeoDelta supervision and used for the attention probes."},{"cited_title":"RoboTwin 2.0: A Scalable Data Generator and Bench- mark with Strong Domain Randomization for Robust Bi- manual Robotic Manipulation","cited_arxiv_id":null,"evidence_quote":"RoboDojo is the benchmark whose object motion is estimated with DA3 depth rescaled by one fitted factor per episode, the site of the paper's depth-scale assumption."},{"cited_title":"Egoscale: Scaling dexterous manipulation with diverse egocentric human data,","cited_arxiv_id":null,"evidence_quote":"VLABench is the benchmark behind the unseen-texture robustness claim, where the margin over OpenWAM-Alpha grows from 0.6 to 7.4 points."},{"cited_title":"OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining","cited_arxiv_id":"2609.07398","evidence_quote":"OpenWAM-Alpha is the strongest RGB-future WAM baseline on all three benchmarks and one of the models probed for action attention."},{"cited_title":"Training-Free Interaction-Aligned Visual Token Pruning for Efficient Embodied Manipulation","cited_arxiv_id":"2603.22991","evidence_quote":"FlowWAM is the RGB-plus-2D-flow baseline on the real robot whose in-distribution lead (75.5%) collapses under visual shifts to 62.0%."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":2}