We introduce 4DCodeBench, a benchmark for 4D inverse graphics through code generation, in which agents reconstruct dynamic scenes from video as executable graphics programs. To accomplish this, agents must translate visual observations into compact representations of scene structure and dynamics, by implementing abstractions such as physical simulations to reproduce complex behavior. To evaluate this capability, we curate a set of real-world videos and construct synthetic scenes spanning diverse physical phenomena, including deformation, fluid flow, and fracture. We perform extensive benchmarking of frontier models, finding that strong static reconstruction capabilities do not yet translate into reliable reconstruction of complex dynamics. 4DCodeBench provides a testbed for tracking progress toward agents that can interpret the dynamics of the world through code. Our benchmark is available at this https URL
https://arxiv.org/abs/2610.03715
We present FlowHMR, a framework for recovering physically plausible global 3D human motion from monocular video. Previous learning-based methods typically regress human motion directly from video and train the network with geometric supervision. However, recovering human motion from monocular video is inherently ambiguous in depth, and direct regression tends to collapse toward an averaged solution. Moreover, the recovered motions are not guaranteed to be physically plausible, so physics-based tracking of them often fails. To address these challenges, we formulate video motion capture as a video-conditioned motion generation problem and first pretrain a flow matching model for this task. Given an input video, the pretrained model generates diverse motion candidates, but not all of them are faithful to the video or physically trackable. We therefore post-train the model using Group Relative Policy Optimization (GRPO) with two rewards. A fidelity reward encourages consistency with the input video. A tracking reward favors motions that a physics-based controller can track successfully. Together, these rewards shift the model's output preference, so the post-trained model stays faithful to the input video while producing more physically plausible motion. We further introduce Wild-4K, a large and diverse dataset of about 4K internet videos, for evaluating human motion recovery in the wild. Qualitative and quantitative experiments on Wild-4K show that our method outperforms state-of-the-art methods in overall motion fidelity and achieves a physical tracking success rate of 82.47%, compared with 62.82% for the strongest baseline, GVHMR.
https://arxiv.org/abs/2610.03691
We present a generalized implementation of Look-Back and Look-Ahead Adaptive Model Predictive Control (LLA-MPC), a learning-free framework for real-time, rapid adaptive system identification and control. The original formulation was demonstrated only in simulation, for autonomous racing with a fixed model structure. Our implementation has modular dynamics and integrators, and we validate it on the F1TENTH platform under constrained computation and noisy state estimation. The system identifies tire parameters online by evaluating thousands of candidate models in real time on an embedded computer. Experiments with low-friction tires across changing surfaces show that LLA-MPC completes high-speed tracking tasks where a fixed nominal model fails. Code, videos, and our related work are available at: this https URL.
https://arxiv.org/abs/2610.03616
A targeted adversarial perturbation can drive a vision-language model's (VLM's) teacher-forced training loss for a fixed target caption to near zero, yet the same model, allowed to generate freely, produces the original, correct description with no trace of the target. We call this dissociation the train/inference gap, and give it a precise mechanistic account on Qwen2.5-VL-7B-Instruct using a controlled two-stage PGD attack on 200 held-out COCO images. First, we show that image-level pixel statistics, including a correctly re-implemented, texture-based attackability measure from the CNN robustness literature, have essentially no predictive power over which images are corrupted (best predictor r=-0.050, p=0.484; ridge regression R^2=0.069). Second, using the logit lens, we localise the gap to a single autoregressive step: the rank of the target token, conditioned on the correct first token already being generated, is fixed at exactly 3,488 out of 152,064 vocabulary entries for every image and every condition, with zero variance. Third, tracking target-token rank across all 28 LLM decoder layers reveals that the visual encoder corrupts every image's representation by a comparable margin regardless of eventual outcome, but the language model decoder then differentially arbitrates: amplifying the corrupted signal for susceptible images and actively suppressing it, past its clean-image baseline, for resistant ones (p<0.001, rank-biserial r=0.579). A linear probe on the merger hidden state separates these two outcomes with AUC=0.858, though we flag a circularity concern in this estimate. Together these results argue that adversarial robustness in autoregressive VLMs is substantially a property of the language decoder's prior, not the visual encoder, with direct implications for where faithfulness evaluations and defenses for deployed VLM systems should be targeted.
https://arxiv.org/abs/2610.03445
Music information retrieval often relates signal-derived, algorithmic, and symbolic information across different coordinate systems. Existing visualizations typically leave notation separate from physical time, while composites that combine them are assembled by hand. We present LayerIt, an open-source Python library for composing independent representations with notation on a shared performance-time axis. Given a score warped into performance time and a note-level alignment, LayerIt emits a single SVG that preserves the score's MEI structure and keeps added components identifiable and restylable. We demonstrate this through a challenging beat-tracking analysis, in which tracker output, signal representations, and notation can be inspected together to locate metrical disagreement and other errors against both notated structure and performed time.
https://arxiv.org/abs/2610.03428
Video is an abundant, inexpensive source of human motion data that is rich in extreme athletic behaviors. Making it usable for humanoid robots, however, is not a matter of simply retargeting a reconstructed trajectory: video-derived motion is physically inconsistent, devoid of actuation information, and says nothing about failure or recovery. We present KungfuAthleteBot (KAB), a framework that treats learning high-dynamic motion from video as the central problem and resolves each of these three failure modes in turn. (C1) We build the KungfuAthlete dataset from videos of national-level martial artists and introduce a physics-guided parabolic trajectory correction that removes height floating, ground penetration, and high-frequency jitter from reconstructed aerial and landing phases. (C2) Because video carries no force information, strict tracking of a reconstructed trajectory is dynamically infeasible, and error-driven initialization keeps re-launching the policy from infeasible aerial poses. We introduce physics-driven pseudo-low-kinetic-energy (LKE) sampling, our central mechanism for making such references learnable: it biases initialization towards dynamically feasible states, letting the policy discover feasible actuation patterns instead of imitating infeasible ones. (C3) Finally, we introduce a direct training paradigm in which disturbance rejection and fall recovery are learned inside the same policy that tracks the video motion, requiring no recovery reference data and no manual mode switching. On a humanoid robot, KAB learns dynamic skills from video and recovers from arbitrary falls in about 0.7 s, the fastest reported recovery for a unified policy. Ablations on the unified policy confirm the necessity of its components, supporting the view that repairing and compensating video data, rather than only collecting more of it, is what unlocks high-dynamic humanoid skills.
https://arxiv.org/abs/2610.03388
The ability of a device to localize itself within its surroundings is a fundamental prerequisite for spatial computing. Visual-inertial odometry (VIO) has proven to be a cost-effective and accurate solution for this task. Robots, wearables, XR devices, and drones can benefit significantly from efficient implementations of VIO since they allow for cooler, lighter, and cheaper devices with longer battery life and a better user experience. In this work, we propose to enhance the efficiency of a VIO system by leveraging the Hexagon DSP, a commodity co-processor present in many modern smartphones and XR devices. Our approach offloads the visual frontend of a stereo-inertial odometry system to the DSP while keeping the backend on the main CPU. By optimizing the implementation for the DSP architecture, we achieve significant reductions in power consumption and latency compared to CPU-only execution. Our system, HexVIO, demonstrates a 67% reduction in power consumption or an 86% increase in throughput on a commodity smartphone, with the ability to sustain long-term real-time 30 fps tracking for 0.83 W, corresponding to ~18 hours of tracking on the testing device. These results highlight the potential of commodity DSPs for enabling all-day visual-inertial tracking in robotics and mobile devices.
https://arxiv.org/abs/2610.03283
Dynamic-point filters are routinely added to feature-based visual SLAM, and several recent systems argue that removing dynamic features can leave too few static features in low-texture regions. So far, these systems have been evaluated only on texture-rich benchmark sequences. We present a controlled study that isolates this interaction. We render synthetic indoor sequences in which surface texture (four levels, quantified by FAST-corner density and image-gradient entropy) and scene dynamics (three levels) are varied factorially along identical camera trajectories, with stereo, RGB-D, ground-truth poses and dynamic masks. On this grid we compare ORB-SLAM2 without filtering, with an optical-flow and epipolar-residual filter (FLOW), and with a multi-view depth-consistency filter (GEOM), and report trajectory error, tracking completeness and surviving static features over five runs. Because the masks give per-keypoint ground truth, we also measure each filter's dynamic-point precision and recall and its static-feature false-removal rate, so that mechanistic explanations can be tested directly. We do not propose a new filter. On 720 runs over 24 sequences, filtering helped mainly in the most dynamic cells; the benefit did not decline monotonically with texture, but at the lowest level filtering reduced tracking completeness, and ORB-SLAM2 never initialised in static L3 scenes. Contrary to our hypothesis, GEOM discarded more static keypoints than FLOW (median FRR 6.8% vs. 1.5% for RGB-D, 19.6% vs. 1.5% for stereo); its RGB-D advantage tracked dynamic-point recall and vanished in stereo mode. Data and code are available at this https URL.
https://arxiv.org/abs/2610.10564
Knowledge-driven neural vocoders struggle to learn reliable fundamental frequency end-to-end, because spectral objectives provide weak supervision of periodic structure and lack phase information. We address this with a source-filter model whose alias-free additive source makes the instantaneous phase of the glottal cycle explicit; differentiating it yields the instantaneous frequency, and thus $F_0$, without an external tracker. Waveform error supervises only the deterministic harmonic path, while a spectral loss covers the full signal. On M4Singer and LM-SSD, the reconstruction is phase-aligned, reaching a signal-to-reconstruction-error ratio of 8.1 dB, while neural baselines remain negative. However, GOLF, given an external $F_0$, still reaches lower spectral distortion. On LM-SSD, the recovered $F_0$ attains the highest overall accuracy of any method tested, including supervised neural pitch trackers applied off the shelf, and the glottal closure instants come within 0.53 points of REAPER's identification rate, without any $F_0$ label.
https://arxiv.org/abs/2610.03058
Social bees are important pollinators that support biodiversity and crop pollination globally and serve as important model systems for collective behavior, but scalable measurement of individual- and colony-level behavior remains difficult in dense, occluded nest environments. Existing monitoring workflows use fiducial tags (e.g., ArUco) to preserve individual identity, yet tag-based tracking can fail when markers are obscured and provide limited information about body extent, spatial context, and untagged individuals. We present BeeWhere, an AI-assisted annotation and analysis workflow that combines ArUco detections with deep-learnt instance segmentations to quantify bumble bee behavior from high-resolution colony images and videos. Using bumble bee (Bombus impatiens) microcolonies as a test case, we annotate 483 frames containing 8,443 bee instances. We additionally annotate pollen balls, nest structures, and chamber boundaries, and train YOLO instance segmentation models for downstream behavioral analysis. Instance segmentations enable quantification of important behavioral metrics based on body contours, including nearest-neighbor distance, proximity to nest structures, spatial occupancy within the nest, and detection counts over time. We apply the BeeWhere models to tag-based tracking in an exploratory validation study assessing the behavioral impacts of neonicotinoid pesticide exposure. BeeWhere increased detection rates compared to tag-based tracking, particularly when bees were partially obscured or under challenging imaging conditions, and also captured treatment-associated changes in bee spatial organization not captured using tag-based tracking alone. These results suggest that instance segmentation can complement fiducial-marker tracking by recovering behaviorally meaningful signals under challenging colony conditions.
https://arxiv.org/abs/2610.03051
During contact-rich manipulation, interactions between a robot and its environment carry information about local geometry: a surface prevents penetration, a bore guides a peg. A controller that exploits these interactions can comply with environmental constraints while preserving task intent. Robot-earning systems commonly use position, hybrid force-position, or Cartesian impedance control. Their prescribed tracking objectives, stiffness, or force-control directions may not match local constraints and may degrade performance. We learn a proprioceptive reflex policy in simulation on three simple interaction primitives: a spring, a plane, and a rail. The policy maps task-space commands to joint-position targets using state history, without direct force or geometrical measurements. Once trained, it serves as a frozen execution layer beneath higher-level controllers. We evaluate it in dual-arm box lifting, peg insertion, and surface following. In box lifting, the reflex kept the force below the threshold while the baseline failed. In rough-surface following it stayed below the 10 N reference, on par with tuned hybrid force-position control. In 0.02 mm peg insertion It reduced mean estimated contact force to less than half that of the baseline while increasing hardware success rates from at most 22 % to 36-58 %. By separating contact response from command generation, the reflex policy provides motion planners, learned policies, and teleoperators with robust contact-rich execution.
https://arxiv.org/abs/2610.02811
Harness evolution improves an LLM agent's prompts, tools, and workflow, while the optimizer's own tools and procedures often remain fixed. We study whether an optimizer can improve another agent more effectively by also improving how it diagnoses failures, develops edits, and tests their effects. Two observations guide our design. In a controlled study, optimizer self-evolution fails to improve performance without execution-based verification, but achieves the best result of that study when verification is available. Across five executors, self-evolving optimizers build their own tools for failure analysis, verification, training audits, and workflow control. Motivated by these findings, we introduce VERSE, a Verified Self-Evolving optimizer for agent harnesses. VERSE lets the optimizer test draft edits, replay failures, and perturb suspected steps before submission, while tracking fixes and regressions across rounds. Using this feedback, the optimizer revises both the executor harness and its own prompts, skills, tools, hooks, and notes, while the weights of the optimizer and executor models stay fixed. Under a shared protocol with disjoint training, validation, and test tasks, VERSE improves all four evaluated harness optimizers on held-out SWE-rebench tasks and newer out-of-distribution tasks in five languages. Its best validation-selected harness reaches 42.3% and 37.7% accuracy, respectively, against 39.2% and 29.3% for the strongest baselines. Code is available at this https URL.
https://arxiv.org/abs/2610.02616
Physical AI Smart Spaces is, to the best of our knowledge, the first benchmark to simultaneously provide large-scale, multi-class, and multi-camera 3D perception data for indoor smart spaces. It contains over 280 hours of synchronized 1080p footage captured by nearly 1,800 cameras in warehouses, hospitals, retail venues, and similar settings, together with automatic annotations for multi-camera identities, 2D bounding boxes, 3D bounding boxes, camera calibration, and depth where available. The benchmark spans Isaac Sim synthetic generation, Cosmos Transfer appearance augmentation, and real-world Sim2Real evaluation. For the real-world target, we include two warehouse deployments with time-synchronized streams, automatic VGGT-based calibration, and a 3D labeling interface that projects world-frame 3D boxes into each view for cross-camera verification. We describe the dataset scope, annotation and calibration schema, generation workflow, benchmark protocols, and official evaluation system, which standardizes submission format, and leaderboard reporting. A central contribution is a 3D instantiation of Higher Order Tracking Accuracy (HOTA), extending the usual 2D box-based tracking evaluation to 3D locations and 3D boxes. We further report empirical baselines from the AI City Challenge leaderboards, showing how methods evolve from person-only 3D location tracking to multi-class 3D box tracking under realistic smart-space constraints. The release is available at this https URL.
https://arxiv.org/abs/2610.02580
Automatic horizon tracking is a foundational task in 3D seismic interpretation. Most existing deep learning approaches formulate it as dense semantic segmentation, typically using U-Net-based architectures. The model produces a probability map over all pixels that must be post-processed to extract precise horizon coordinates, while horizon picks in time/depth must be converted into dense masks for training. Unpicked seismic traces are consequently treated as background, which can hinder convergence, and both pre- and post-processing can introduce errors into the final interpretation. Moreover, 2D segmentation models do not inherently capture inter-slice context, while 3D models are often computationally prohibitive. We instead formulate horizon tracking as a bounded coordinate regression problem, where the model directly predicts the time/depth coordinate of the target horizon at each lateral position. We propose a lightweight regression head compatible with any pretrained vision backbone, coupled with an LSTM module to model inter-slice context and produce a continuous horizon surface across the volume. A combination of L1 and L2 losses supervises predictions at valid horizon picks, while a geology-informed regularization enforces lateral continuity between successive traces. Under controlled experimental conditions, we evaluate four pretrained vision backbones under both segmentation and regression configurations on a seismic volume from New Zealand. The proposed approach consistently outperforms its segmentation counterparts quantitatively, using metrics including RMSE and PCC, and qualitatively, while also demonstrating greater robustness to increasing sparsity of training picks. Finally, we show that prediction variation across successive traces captures local variations in geological complexity, providing an automated quality control measure for downstream seismic interpretation.
https://arxiv.org/abs/2610.02494
Geometric understanding supports endoscopic navigation and robotic assistance, but learning reliable endoscopic geometry faces two challenges: scarce geometric annotations and ambiguity between camera motion and tissue deformation. We present SCOPE-4D, an endoscopic 4D geometry foundation model that jointly predicts camera parameters, dense geometry, and 3D tissue trajectories from monocular RGB video in a single forward pass. Our curation and annotation pipeline constructs SCOPE-5K, a collection of approximately 5,000 clips spanning real and synthetic gastrointestinal endoscopy and laparoscopy. The collection provides rich geometric supervision and includes newly collected phantom and real-colonoscopy evaluation sets. Geometric supervised fine-tuning on SCOPE-5K learns endoscopic priors that improve camera and depth estimation. Common--Residual Motion (CRM) further constrains local deformation relative to common tissue movement. Together with geometric supervision, CRM and trajectory supervision further improve camera and depth estimation over geometric fine-tuning alone while enabling dense 3D tissue tracking. Evaluations on public and newly collected benchmarks demonstrate strong in-domain and out-of-domain geometry, superior 3D tracking, and more stable long-sequence colon reconstruction. A blinded user study further supports the perceived reconstruction quality on real clinical video. Together, these results demonstrate the value of large-scale endoscopic supervision and motion constraints for joint geometry estimation and tissue tracking.
https://arxiv.org/abs/2610.02343
Safe whole-body motion is essential for deploying humanoid robots in unstructured environments. Modern humanoid control commonly separates reference specification from execution, with a planner, teleoperator, or motion generator providing a reference that a reinforcement-learning policy tracks through dynamically feasible whole-body control. Runtime safety filters, such as control barrier functions (CBFs), offer a promising approach for enforcing newly introduced constraints via interventions on the tracker's outputs. We show, however, that treating the tracking policy and safety filter independently induces fundamental mismatches, as filtering alters both the executed actions and the induced state distribution. We study this policy-filter interface through case studies that isolate dynamics, objective, and information mismatches, highlight their root causes, and use these insights to develop CoFiT (Constrained Filter-aware Tuning), a filter-aware fine-tuning method for pretrained trackers. Across diverse constraint scenes, CoFiT reduces violation time relative to filter-only training by 91% on TWIST2 and 21% on SONIC, while requiring smaller safety filter corrections. On Unitree G1 hardware, CoFiT reduces violation time by 83% for TWIST2 and completes every trial without operator intervention, whereas 50% of baseline trials require an operator stop. Together, these results provide actionable insights into policy-filter interactions and establish design principles for integrating learned trackers with runtime safety filters.
https://arxiv.org/abs/2610.02341
We introduce Triangular Resampling (TR), a post-training method for mitigating long-horizon error accumulation in motion diffusion models. Built on FloodDiffusion's triangular denoising schedule, TR addresses the mismatch between ground-truth-derived training windows and model-generated inference states. Replacing only completed motion history leaves this mismatch unresolved in partially denoised states within the active window. TR therefore extends rollout-based training to these states, using ground-truth clamping to limit excessive drift. For each replayed sample, TR draws one denoising threshold, shared across latent positions and replay updates, and replays multi-step triangular denoising without gradient tracking. After each update, states below the threshold are replaced with noise-matched ground truth, while those at or above it retain model predictions. The resulting latent window enters the standard training update. This rollout construction supports both supervised training (TR) and distribution matching (TR-DMD). On 120-second motion generation from HumanML3D test prompts, TR and TR-DMD achieve state-of-the-art FID AUC within their respective non-DMD and DMD comparison groups. Supervised TR reduces FID AUC by 40.9% and FID degradation slope by 55.3% relative to matched post-training without replay.
https://arxiv.org/abs/2609.34697
Spatial reasoning benchmarks evaluate vision-language models across diverse tasks, but task-level scores do not reveal which underlying capabilities account for success or failure. Each task requires recovering spatial evidence, representing geometry, and reasoning over it. We disentangle these capabilities by comparing predicted and ground-truth spatial context under a shared schema and coordinate contract. This comparison reveals four recurring sources of error: inaccurate perception, missing information in the spatial context, selection of the wrong measurement, and errors in reference frames or in tracking position and orientation. Guided by this diagnosis, we develop CROSS, a training-free library of typed geometric operators and spatial skills that function over available evidence to support reliable video spatial reasoning. The resulting library supplies verified context to non-coding VLMs or callable skills to a SpatialClaw agent. We evaluate \methodname{} on five benchmarks. \methodname{} raises the average score from 55.9\% to 60.2\% on ReVSI and improves the SpatialClaw result from 62.8\% to 66.3\% on DSI-Bench. These gains demonstrate that explicit handling of spatial conventions can repair systematic reasoning failures without additional training.
https://arxiv.org/abs/2610.01999
Feedforward foundation models have recently shown remarkable 3D reconstruction capabilities. However, existing models exhibit large tracking drift in long-context streaming reconstruction due to error accumulation. In this paper, we revisit loop closure with streaming reconstruction foundation models to enable accurate, drift-free, kilometer-scale reconstruction. Specifically, our method detects loop candidates through global descriptor retrieval, and constructs loop-conditioned windows to estimate the relative poses between looped frames. Given the observation that our adopted streaming reconstruction backbone produces a globally consistent scale, we optimize all frame poses on the SE(3) manifold with sequential and loop closure constraints, avoiding the pose graph optimization on the Sim(3) or higher-dimensional SL(4) manifolds employed in prior works. Extensive experiments show that our method reduces drift and produces consistent geometry on kilometer-scale sequences, significantly outperforming the state of the art. Code is available at this https URL.
https://arxiv.org/abs/2610.01927
Human hand-object demonstrations offer a reusable source of dexterous robot manipulation data, but transferring them across embodiments requires physically feasible retargeting. Existing physics-based approaches face limitations in retargeting success, motion-specific training efficiency, or both. To address these limitations, we introduce FlashDexRetarget, an RL-based framework for high-success, efficient dexterous motion retargeting. To make the demonstrated interaction easier to learn, we combine object point-cloud observations, hand-object distance features, and future trajectory encodings with complementary rewards that supervise object motion and reference hand-object relationships. To further accelerate learning, we employ separate left- and right-hand actor critic networks and adapt the off-policy algorithm, FlashSAC to dexterous motion tracking. On a benchmark of 50 motions spanning single-object and two-object interactions, FlashDexRetarget achieves a 90% success rate, approximately 2.5x that of the evaluated sampling-based baselines, while requiring up to 100x less training compute than the evaluated RL-based baselines. Evaluations on both XHand and Sharpa Wave Hand show consistent gains, and component-wise ablations examine the contributions of our design choices. Beyond the 50-motion benchmark, experiments with 200, 500, and 1,000 motions demonstrate that our method remains stable at larger scales and produces successful retargeted motions more efficiently as the training set grows. Qualitative replay results using real-world-captured demonstrations further illustrate the applicability of our framework to recorded human manipulation. Videos and code are available at this https URL
https://arxiv.org/abs/2610.01849