A new method for informed sampling on Riemannian manifolds uses Loewner order lower bounds to produce tighter informed sets, accelerating motion planning for robotic manipulators.

Researchers develop a digital twin system for hydroponic lettuce, using sensors and neural networks to track individual plant growth and predict yield. The custom neural network estimates mass within 1.5 g from RGB-D images, and the integrated system forecasts yield 1-4 days ahead with ~2 g error.

A self-supervised Hybrid Adaptive Kalman Filter is proposed that learns structured corrections to system dynamics and noise covariance from measurements only, enabling probabilistic model classification via innovation likelihood. Experiments show improved estimation accuracy and robust classification in both low- and large-data regimes.

SeeTraceAct is a demo-conditioned VLA framework that improves spatial grounding via visibility-aware prediction of future end-effector traces. It outperforms baselines on RoboCasa-DC and real-world tasks, boosting average success by 12.5 percentage points.

This paper introduces an estimation and control framework for dynamic landing of multi-rotor uncrewed aerial vehicles on moving platforms. The proposed method integrates nonlinear model predictive control with a real-time minimum-jerk trajectory planner that enforces a prescribed touchdown time, enabling consistent timing during the terminal descent. To enhance robustness in the presence of time-varying sensing quality, we utilize an adaptive unscented kalman filter that updates the process and measurement noise statistics online. In addition, we provide a reference feasibility analysis showing that minimum-jerk references induce bounded thrust and torque commands under standard tracking hypotheses. The proposed framework is evaluated in simulation and hardware experiments, and it is show…

CARVE is a prediction-free certificate layer for interactive driving that repairs vetoed maneuvers by identifying bounded multi-agent edits when hard-rule margins are negative. On 589 INTERACTION replay episodes, CARVE-Greedy accepts 98.64% of initially vetoed maneuvers and recovers 370/378 human-resolved false vetoes, while preserving zero priority-agent false positives and 400/400 negative-stress vetoes. CARVE does not rely on prediction; it certifies whether an interaction is bounded, attributable, and normatively admissible under declared assumptions.

While sim2real efforts are necessary for effective policy transfer to hardware, there is such a thing as too much of a good thing. We argue that sim2real efforts have led to misaligned incentives with policy learning, resulting in simulator lock in and poor policy exploration due to the unreasonable constraints imposed by the real world. We offer a diagnosis and explanation of the current status of the problem, and propose a potential solution via a sim2sim2real paradigm that leverages the robot's kinematics as the sole design constraint.

This paper presents an automated agent-driven pipeline that generates multiple-choice VQA datasets from paired private radiology reports and 3D oncology imaging, producing two complementary question types: RADS-style and report-derived questions. Evaluated on four in-house cancer cohorts, zero-shot evaluation reveals no dominant model and substantial headroom. A blind ablation shows visual reliance is dataset-specific; lung CT is solvable without images. The pipeline is released as an open agent skill.

NVIDIA introduces Cosmos 3, a family of omnimodal world models that jointly process and generate language, image, video, audio, and action sequences using a unified mixture-of-transformers architecture. It achieves state-of-the-art on understanding and generation tasks, and is released open-source under the OpenMDW-1.1 license.

Human-object interaction (HOI) recognition is critical for automatically analyzing student behavior in complex educational environments. Although state-of-the-art HOI detectors perform well on benchmark datasets, their performance often degrades in real-world training environments due to domain-specific objects, occlusions, and complex visual conditions. This paper introduces a diagnosis-driven framework that integrates a triplet-level HOI error taxonomy with error-factor attribution analysis for real-world educational video data, studied in the context of Critical Care Air Transport Team (CCATT) mixed-reality medical training. The approach improves the macro-F1 score of a pretrained CDN model from 48.6 to 90.2 on the CCATT dataset.

GeoDrive-Bench is a new benchmark for evaluating vision-language models on region-specific traffic rules for autonomous driving. It contains 5,053 human-validated multiple-choice questions from six countries, covering perception, prediction, planning, and region reasoning. The paper also proposes a distillation algorithm to inject local traffic knowledge into VLMs, showing that current VLMs lack robust region-aware reasoning, but the method improves cross-region performance.

This study evaluates Random Forest and four CNNs (ResNet-50, ResNet-101, EfficientNet-B4, ConvNeXt-Large) for transferable satellite-derived bathymetry over 0-20 m depth using Sentinel-2 imagery. Key design choices include preserving spatial continuity (contiguous reef blocks) and a Smooth Weight Function (SWF)-weighted RMSE loss. Intra-regional RMSE ranges from 1.15-1.92 m (as low as 0.26 m for shallow depths), while cross-regional RMSE is 2.46-2.98 m for deep models. On the MagicBathyNet benchmark, the proposed networks achieve 0.19-0.22 m RMSE, outperforming U-Net and a task-specific transformer with fewer parameters. Multi-temporal imagery and median aggregation reduce noise. Optimized architectures and pretrained weights are released for scalable transfer.