{"id":"5eb8933e-f3f4-4954-b09d-107581af320b","arxiv_id":"2609.13812","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"GeomVLA achieves state-of-the-art robot manipulation by integrating latent 3D scene motion prediction into a vision-language-action model with a shared metric coordinate frame.","lead":"GeomVLA unifies perception, future scene motion prediction, and robot action generation within a shared 3D coordinate frame. This consistent geometric representation improves robot manipulation performance in simulation and real-world tasks without requiring large-scale robot-action pretraining.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"No significant objection identified","rationale":"The reader's weakest_assumption (pseudo-label accuracy) is technically correct but does not constitute a load-bearing concern for the central claim. The paper provides converging evidence that the motion representation is meaningful and beneficial, independent of the absolute accuracy of the pseudo-labels: instruction sensitivity (Table 7), horizon impact (Table 6), the parameter-count control, and the systematic ablation pattern all support the claim that geometric consistency is the primary driver of improvement. The RoboTwin2.0 training-set confound is acknowledged and does not affect the core ablation results on CALVIN and LIBERO, which are consistent across seeds. Therefore, the present stress-test identifies no objection that would warrant a change to the ACCEPT verdict.","tokens_in":20936,"tokens_out":6215,"duration_ms":55418,"concrete_test":"Independently evaluate the 3D Scene Trajectory Denoiser's pseudo-labels by comparing with ground-truth trajectories from a motion-capture system on a subset of real-world tasks (e.g., Insert marker, Stack blocks), then retrain the GeomVLA CALVIN model with the corrected labels and recompute the full ablation table (Table 2) to verify that the performance gain from geometric consistency persists under verified supervision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After careful review, the central claim—that maintaining geometric consistency among scene representation, motion prediction, and robot actions is the primary source of improvement—is well-supported by the ablation study (Table 2). The pseudo-label reliance (SpatialTrackerV2, §3.2) is a common limitation, but the motion representation's utility is confirmed through multiple controls: instruction sensitivity (Table 7), performance degradation when motion is removed or represented in 2D (Tables 1–2), and the parameter-count control (random init, Appendix). The RoboTwin2.0 five-task comparison (Table 1) uses different training sets, but this is transparently stated and does not affect the central ablation results on CALVIN and LIBERO, which are consistent and reproducible across two seeds. No fundamental flaw, inconsistency, or hidden assumption that would overturn the central claim was found.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper proposes GeomVLA, a Vision-Language-Action model that unifies perception, latent 3D scene-motion prediction, and action generation within a shared robot-centric 3D coordinate frame. It lifts pretrained VLM features into 3D scene tokens using depth and calibration, introduces a 3D Scene Trajectory Denoiser trained via pseudo-labels to predict future point motion, and conditions a flow-based 3D action denoiser on intermediate motion features through geometry-aware attention. The method achieves state-of-the-art performance on CALVIN (4.624 mean tasks), competitive performance on LIBERO (98.4%) and RoboTwin2.0 (84.8% on a five-task subset), and outperforms baselines in real-world manipulation without robot-action pretraining. A thorough ablation study supports the central claim that maintaining geometric consistency among scene, motion, and action representations is the primary source of improvement.","tokens_in":21150,"tokens_out":1616,"duration_ms":16318,"significance":"If the claimed results hold, GeomVLA makes a solid contribution to the VLA literature by demonstrating that explicit 3D geometric consistency across perception, intermediate motion reasoning, and action control yields clear gains over both 2D alternatives and motion-only baselines. The ablation study is well-designed and isolates the effect of shared geometry from other factors. The method achieves competitive or state-of-the-art results on multiple benchmarks without large-scale robot-action pretraining, and the real-world evaluation adds practical credibility. The paper is clearly written and the architectural choices are well motivated.","major_comments":[],"minor_comments":[{"comment":"The motion-weighted loss uses weights wi,t that depend on a soft motion-saliency term. The main text describes the weight formula only briefly in the appendix. Consider adding a short summary (e.g., the motivation for upweighting moving points while retaining static structure) to the main exposition, as the loss design affects the quality of learned motion features.","section":"§3.2, Eq. (2)"},{"comment":"The footnote explains that the five-task comparison uses different training sets (GeomVLA trained on all 50 tasks, while GeomVLA w/o motion and the joint-angle variants are trained only on five tasks). This is transparent but easily missed. Moving the caveat to the main caption or adding a note in the main text would improve clarity.","section":"Table 1, RoboTwin2.0 column"},{"comment":"The number of demonstration rollouts varies (50 for three tasks, 20 for five). A brief rationale for this split (e.g., task difficulty, data collection budget) would help readers assess the fairness of the comparison across tasks.","section":"§4.3, Real-World Experiments"},{"comment":"The trajectory denoiser hidden dimension for the real-world experiments is 128, notably smaller than for simulation (320–512). It would be helpful to comment on whether this was due to dataset size or computational constraints, and whether the final results might be sensitive to this architectural choice.","section":"Appendix, Table 4"}],"recommendation":"accept"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"This is a well-executed paper that delivers on its central claim: maintaining geometric consistency among scene representation, motion prediction, and robot actions in a shared 3D coordinate frame is the primary source of improvement, not future-motion reasoning alone. The architecture is clean—a 3D Scene Trajectory Denoiser feeding intermediate motion tokens into a 3D action denoiser via geometry-aware attention—and the ablations (Table 2) are the strongest part of the paper. They isolate the effect of 3D motion reasoning, the choice of motion token (early-flow beats full trajectory or encoder-only), the importance of a common frame (arbitrary rigid preserves performance; mixing 2D motion with 3D action hurts), and rule out a parameter-count confound. The two-seed stability check on CALVIN and LIBERO is responsible.\n\nThe empirical results are competitive: SOTA on CALVIN, strong on LIBERO and RoboTwin2.0, and consistent real-world gains across eight tasks with 20–50 trials each. The real-world setup with ALOHA2 and multi-view Kinects is realistic. The inclusion of a joint-angle variant (GeomVLA-JA) that still outperforms the 2D counterpart is a nice extra validation that the benefit lives in the 3D representation, not in the action parameterization.\n\nSoft spots are modest and mostly acknowledged. The trajectory denoiser relies on SpatialTrackerV2 pseudo-labels without independent verification, though the motion representation's utility is confirmed through instruction-sensitivity tests (Table 7) and the parameter-count control. Real-world results are on 20 trials per task; some tasks (rank by color, place in triangle) show low absolute success, which the authors honestly attribute to cross-task interference. No code is released (common but notable). The five-task RoboTwin2.0 comparison uses different training sets, but this is transparently stated and the central ablations on CALVIN and LIBERO are fully controlled.\n\nThe paper is clearly written and the claims are proportionate to the evidence. It deserves a serious referee and will be a useful reference for anyone building 3D-aware VLA models.","headline":"GeomVLA makes a solid case that 3D geometric consistency across perception, motion, and action improves VLA performance, with thorough ablations backing the claim.","tokens_in":21692,"tokens_out":1347,"duration_ms":13525},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"GeomVLA unifies scene representation, future-motion reasoning, and action generation in a shared 3D coordinate frame, achieving state-of-the-art performance on CALVIN and competitive results on LIBERO, RoboTwin2.0, and real-world tasks…","keywords":["vision-language-action model","3D manipulation","scene flow","future motion prediction","geometric consistency","flow matching","robot control","point trajectory"],"falsifier":"Train GeomVLA on a large dataset where ground-truth 3D point trajectories are available (e.g., from a motion-capture system or simulator) and compare performance to the variant trained with SpatialTrackerV2 pseudo-labels. If the pseudo-label variant matches or exceeds the ground-truth variant, the assumption that label quality is critical is weakened; if it is significantly worse, the reliance on pseudo-labels is a bottleneck.","tokens_in":22530,"feed_emoji":"🤖","tokens_out":2500,"duration_ms":23336,"temperature":0.7,"pith_summary":"The paper argues that vision-language-action (VLA) models for robot manipulation benefit when scene perception, latent motion prediction, and action control all operate in the same metric 3D space. Most prior VLAs either work in 2D or decouple these components, losing geometric consistency. GeomVLA lifts pretrained VLM features into 3D scene tokens using depth and camera calibration, introduces a task-conditioned 3D Scene Trajectory Denoiser that predicts future point motion, and feeds intermediate motion features into a 3D flow-based action denoiser via geometry-aware attention. Experiments show that maintaining a shared robot-centric 3D frame throughout the pipeline yields larger gains than future-motion reasoning alone, across simulation benchmarks and real-world tabletop tasks.","feed_headline":"Shared 3D frame unifies perception, motion, and action for robots","feed_subtitle":"GeomVLA predicts future scene motion in metric space and uses it to guide manipulation, outperforming 2D reasoning and decoupled approaches.","key_machinery":"The central mechanism is the 3D Scene Trajectory Denoiser, a flow-matching module that predicts per-step 3D displacement increments for a grid of image points lifted into the robot base frame. Its intermediate decoder state (at initial noise, not the final trajectory) is pooled into motion tokens and fused with scene tokens via a geometry-aware cross-attention layer, which then conditions a separate 3D rectified-flow action denoiser that outputs end-effector poses. Positions, predicted displacements, and action tokens all reside in the same robot-centric metric space.","core_discovery":"GeomVLA demonstrates that explicitly linking scene representation, future point-motion prediction, and action generation within a single 3D coordinate frame improves language-conditioned robot manipulation performance. The key finding is that geometric consistency across these three stages—not just predicting future motion in any representation—is responsible for the gains. Ablations show that using 2D motion features with a 3D action denoiser hurts performance, and that applying the same arbitrary rigid transformation to all three components preserves most of the benefit, confirming the primacy of shared geometry.","pith_inferences":["The performance gap between GeomVLA and the variant using 2D motion with 3D actions implies that aligning the geometric domain of intermediate reasoning with the action space is at least as important as the reasoning itself.","Since GeomVLA is trained only on benchmark-scale demonstrations without large-scale robot-action pretraining, the shared 3D design may offer a path to data efficiency that could be amplified with more diverse data.","The 3D Scene Trajectory Denoiser acts as a learned motion prior that is task-conditioned; this suggests that combining geometric motion priors with language instructions could be a general principle for long-horizon manipulation."],"forward_implications":["Future VLA architectures should embed perception, reasoning, and control in a common 3D coordinate system rather than treating them as separate modalities.","Intermediate motion features from an early stage of the trajectory denoiser are more effective than final denoised trajectories for conditioning the action policy.","The method's reliance on pseudo-labels from SpatialTrackerV2 suggests that high-quality 3D point tracking from video can serve as cheap supervision for motion prediction in robotics.","GeomVLA's architecture supports both Cartesian end-effector control and joint-angle control by feeding intermediate Cartesian action features into a downstream joint-angle denoiser, showing that the 3D geometric representation benefits multiple action spaces."],"supporting_citations":[{"why":"The pretrained Florence-2 VLM provides the initial 2D visual and language features that GeomVLA lifts into 3D.","marker":"[6]"},{"why":"SpatialTrackerV2 generates the pseudo-labels used to train the 3D Scene Trajectory Denoiser.","marker":"[55]"},{"why":"π0.5 serves as a primary baseline for GeomVLA's performance comparisons and is also fine-tuned for RoboTwin2.0 experiments.","marker":"[3]"},{"why":"π0 provides the flow-matching action formulation that GeomVLA adapts and extends with 3D geometry and motion conditioning.","marker":"[2]"},{"why":"ACT3D is an earlier 3D manipulation policy using feature field transformers, which GeomVLA builds upon and compares against in related work.","marker":"[9]"},{"why":"The CALVIN benchmark is one of the main evaluation environments where GeomVLA reports state-of-the-art results.","marker":"[27]"}],"fun_headline_variants":["Robot model unifies scene, motion, action via shared 3D frame","Shared 3D geometry, not future motion alone, boosts robot control","GeomVLA ties scene motion and action in one 3D coordinate frame","Geometry across perception and action matters more than motion foresight","Future motion helps robots only when 3D geometry stays consistent"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The 3D Scene Trajectory Denoiser is trained using pseudo-labels from an external point-tracker (SpatialTrackerV2); if those labels are systematically inaccurate or biased, the learned motion representation and downstream action policy may be degraded.","fun_headline_variants_meta":{"raw":{"variants":["Robot model unifies scene, motion, action via shared 3D frame","Shared 3D geometry, not future motion alone, boosts robot control","GeomVLA ties scene motion and action in one 3D coordinate frame","Geometry across perception and action matters more than motion foresight","Future motion helps robots only when 3D geometry stays consistent"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000219,"raw_usage":{"cost":0.00021854,"is_byok":false,"cost_details":{"upstream_inference_cost":0.00021854,"upstream_inference_prompt_cost":0.00014798,"upstream_inference_completions_cost":7.056e-05},"total_tokens":1309,"prompt_tokens":1057,"completion_tokens":252,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":172}},"tokens_in":1057,"tokens_out":252,"cost_source":"openrouter_usage","duration_ms":2838,"image_pages":[],"temperature":1.0,"reasoning_tokens":172,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-09-15T21:16:45.515209+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":"Train GeomVLA on a large dataset where ground-truth 3D point trajectories are available (e.g., from a motion-capture system or simulator) and compare performance to the variant trained with SpatialTrackerV2 pseudo-labels. If the pseudo-label variant matches or exceeds the ground-truth variant, the assumption that label quality is critical is weakened; if it is significantly worse, the reliance on pseudo-labels is a bottleneck.","supporting_citations":[{"cited_title":"Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks","cited_arxiv_id":"2311.06242","evidence_quote":"The pretrained Florence-2 VLM provides the initial 2D visual and language features that GeomVLA lifts into 3D."},{"cited_title":"SpatialTrackerV2: 3D Point Tracking Made Easy","cited_arxiv_id":"2507.12462","evidence_quote":"SpatialTrackerV2 generates the pseudo-labels used to train the 3D Scene Trajectory Denoiser."},{"cited_title":"$\\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization","cited_arxiv_id":"2504.16054","evidence_quote":"π0.5 serves as a primary baseline for GeomVLA's performance comparisons and is also fine-tuned for RoboTwin2.0 experiments."},{"cited_title":"$\\pi_0$: A Vision-Language-Action Flow Model for General Robot Control","cited_arxiv_id":"2410.24164","evidence_quote":"π0 provides the flow-matching action formulation that GeomVLA adapts and extends with 3D geometry and motion conditioning."},{"cited_title":"Gervet, Z","cited_arxiv_id":null,"evidence_quote":"ACT3D is an earlier 3D manipulation policy using feature field transformers, which GeomVLA builds upon and compares against in related work."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The CALVIN benchmark is one of the main evaluation environments where GeomVLA reports state-of-the-art results."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}