#01Nov 3, 2026
cs.RO
Hybrid Particle-Voxel Dynamics for Multimodal 3D World Simulation
Jimmy Wang, Yifan Chou, Kaifeng Zhang and 5 more
Predicting how the world evolves in response to robot actions is central to policy learning, evaluation, and planning. However, many action-conditioned world models predict future observations in 2D image or video space without explicitly modeling the underlying 3D scene, making spatial consistency across viewpoints and over long-horizon interactions challenging. We introduce 3D World Simulator, an action-conditioned multimodal world modeling framework that jointly predicts appearance, geometry, and tactile observations from a unified 3D representation. Our approach combines a hybrid particle-voxel autoencoder that compresses point clouds into compact 3D voxel latents with an action-conditioned latent dynamics model that predicts their evolution. The shared 3D representation supports geometrically consistent rendering from arbitrary viewpoints and flexible integration of spatially aligned tactile signals. Our experiments demonstrate interactive, long-horizon, multimodal prediction at 50 Hz. We then show how our model enables scalable data generation for policy training and evaluation using tactile and multi-view visual observations, including wrist-mounted and novel camera views. Furthermore, we scale our method to large datasets containing diverse objects and demonstrate model-based planning with unseen objects and robot configurations. Together, these results highlight the potential of hybrid particle-voxel dynamics as a foundation for multimodal 3D world simulation.
#02Nov 3, 2026
cs.CV
A Universal Raw-IQ Radar Encoder for Cross-Sensor Human Sensing
Jonas Leo Mueller, Dario Zanca, Sebastian Hoefler and 5 more
Radar-based human sensing typically relies on end-to-end task-specific models trained on hand-chosen signal projections, which generalize poorly across unseen users, tasks, and hardware. We argue that the dominant signal processing choice, the micro-Doppler spectrogram, is a poor starting point for transfer because it keeps the Doppler axis, which carrier frequency rescales, and discards the angular axis, which carrier frequency leaves unchanged. We instead pretrain a convolutional beta-VAE without labels on clutter-removed raw radar measurements from four platforms spanning 60 and 77 GHz and 4 to 12 virtual antennas, aligned by a unified spectral interpolation grid. For every downstream benchmark the encoder stays frozen, and only a small supervised temporal classifier is trained on its embeddings. We compare against supervised models trained end to end on each benchmark and a pretrained radar foundation model across ten settings covering activity, gesture, sign-language, and gait recognition. When the test users and the radar are seen during training, our encoder is best on all five benchmarks. For users never seen in training, it leads on three of four benchmarks, by 4.8 to 23.2 points. On subject-disjoint gait recognition, it is best for each of the four held-out subjects and 8.3 points above the strongest baseline on average, even though its pretraining excluded every recording of this radar. Excluding the test radar from pretraining costs at most 5.3 points on any benchmark. Code is available at https://github.com/jonasmueler/UniversalRadarEncoder_WACV.
#03Nov 3, 2026
physics.optics
LIPS-PC: Learning Illumination Patterns for Single-Pixel Classification
Tomer Porat, Talya Eden, Eliahu Cohen
In single-pixel (ghost) imaging, known illumination patterns probe an object and a detector with no spatial resolution returns one number per pattern. Each measurement costs acquisition time and photons and, for ionizing probes, contributes to delivered dose. Motivated by automated radiographic defect screening, where a class label suffices, we introduce Learning Illumination Patterns for Single-Pixel Classification (LIPS-PC). The patterns are learned end-to-end while the measurement operator remains one fixed linear matrix, identical for every input. We parameterize it as a residual correction to a two-dimensional discrete cosine transform (2D-DCT), $D=D_{\mathrm{base}}+Δ$, and jointly train the correction and a compact classifier using cross-entropy, without a reconstruction objective. This yields a task-driven linear embedding through optimization. We evaluate LIPS-PC numerically on MNIST and Fashion-MNIST under a controlled protocol, then on RIAWELC, a real industrial X-ray weld-radiograph dataset. Across five budgets, $M\in{8,16,32,64,128}$, corresponding to sampling ratios from below 1% to 12.5%, the learned operator outperforms the best tested fixed transform by 9-15 percentage points at $M=8$ on all three datasets. Its eight-measurement accuracy requires roughly 16-32 measurements with the best fixed transform, indicating a $2\times$ to $4\times$ saving at comparable accuracy. The advantage generally shrinks as $M$ increases, with a small reversal on RIAWELC at higher budgets. These results show that learning what to measure is especially valuable when measurements are scarce.
#04Nov 3, 2026
cs.CV
Gradient-Guided Orientation Learning for Weakly Supervised Oriented Object Detection
Junyong Jang, Juheon Park, Minyoung Jeon and 1 more
Weakly supervised oriented object detection (WS-OOD) aims to reduce the annotation cost of rotated bounding boxes by leveraging cheaper annotations such as horizontal boxes or points. However, these weak annotations provide no explicit orientation information, forcing existing methods to infer object angles through indirect geometric consistency or equivariant constraints. Such indirect supervision often becomes ambiguous and unstable under real-world appearance variations, including illumination changes, color shifts, shadows, and background clutter. In this paper, we present a simple yet effective gradient-guided orientation learning strategy for WS-OOD. Motivated by the observation that the input image itself contains useful orientation priors, our method extracts dominant gradient-based orientation cues from object regions and converts them into pseudo-angle labels for auxiliary supervision, enabling the detector to learn orientation-aware representations without requiring rotated bounding box annotations or extra learnable parameters. Furthermore, since AP can mask substantial angular errors when spatial IoU remains high, we introduce an orientation-aware evaluation metric that incorporates angular discrepancies into the standard AP protocol. Extensive experiments on weakly supervised oriented detection benchmarks demonstrate that the proposed pseudo-angle supervision improves both detection performance and orientation estimation quality, while the proposed metric provides a more faithful assessment of angle prediction than conventional AP alone.
#05Nov 3, 2026
cs.CV
Predicting Void Points in Self-Occluding Forest Volumes
Benjamin Bachmayr, Mohamed Youssef, Oliver Bimber
Synthetic Aperture Imaging (SAI) enables drones to observe forest canopies by warping and averaging multiple displaced views at a common depth. However, occlusion is suppressed rather than eliminated, causing poorly observed regions to appear occupied and leading to systematic biomass overestimation. The Deep Forest project cannot distinguish reconstruction artifacts from real vegetation because its training data relies on measured reflectance points, excluding empty space and preventing the network from learning explicit occupancy boundaries. This thesis addresses the problem by modifying training labels rather than the imaging process. Each simulated scene is rendered twice: as a 9x9 grid of top-down captures for focal-stack generation and as 550 depth-clipped slabs derived from scene geometry to provide occupancy labels, explicitly representing void space as a negative class. A Gazebo pipeline generates procedural forest stands containing 80-200 trees across three species. A pretrained Pix2Pix model translates rendered RGB images into a synthetic near-infrared band, and Light Field Rendering produces aligned focal stacks of 550 planes across three channels. Four architectures were evaluated on a held-out scene. The plain 2D U-Net achieved an IoU of 0.6290, outperforming the volumetric 3D U-Net (0.5707). Established reconstruction methods performed substantially worse: COLMAP achieved 0.3446 and DepthAnything3 0.2959, both after their respective optimal dilation sweeps. An RGB-input variant of the volumetric architecture achieved the highest IoU of 0.6514 without dilation or per-scene calibration. These results highlight the limitations of conventional reconstruction methods, which recover visible surfaces but struggle to infer occupancy within closed canopies.