This paper studies the dynamic robot allocation problem in heterogeneous multi-team collaboration, treating robots as transferable resources, drawing on Hamilton's rule from ecology to design an altruistic decision-making mechanism, and proving that the generalized heterogeneous resource allocation problem is combinatorial optimization and NP-hard.
This paper proposes a feedback world model that uses the true states observed after robot execution to perform online closed-loop correction of predictions at inference time, without additional training data or parameter updates. The method can be interpreted as a latent-space observer and provides convergence guarantees under bounded, asymptotically convergent dynamics residuals and a stable feedback gain; it also introduces action-aware guidance to better translate corrected predictions into control.
Move-Then-Operate is a vision-language-action (VLA) framework that explicitly decouples robotic manipulation into two behavioral phases: coarse relocation (move) and contact-critical interaction (operate), using a dual-expert policy routed by a learnable phase selector.
WARPED is a framework that synthesizes realistic wrist-view observations and actions from human demonstration videos, enabling visuomotor policy training with only monocular RGB data. The system uses vision foundation models to initialize the interactive scene, employs a hand-object interaction pipeline to track the hand and manipulated object and retarget trajectories to a robotic end-effector, and finally uses Gaussian Splatting to synthesize photorealistic wrist-view observations for directly training policies.
Safe-Night VLA is a multimodal manipulation framework that integrates long-wave infrared thermal perception into a pre-trained vision-language backbone, enabling robots to perform semantic reasoning based on thermodynamic properties and to enforce deterministic workspace constraints during policy execution through a control barrier functions safety filter.
ExtremControl proposes a low-latency whole-body control framework for humanoid robots that reduces end-to-end teleoperation latency from about 200ms to as low as 50ms. Its core is adding scaled velocity feedforward in low-level control, which cuts tracking delay by about 70ms, and operating directly on the SE(3) poses of selected rigid links to avoid joint-space retargeting; the paper also derives a closed-form Cartesian-space mapping that needs only a single anthropomorphic calibration to adapt to operators of different body proportions.
The paper proposes an onboard state estimation method for high-speed UAVs using only a monocular RGB camera and an IMU, fusing visual-inertial odometry (VIO), onboard landmark camera measurements, and IMU data, and estimating and compensating for VIO drift through a new mathematical drift model.
HEHA (Hierarchical Exploration with Heterogeneous Agents) proposes a hierarchical framework that uses a new routing algorithm, RCH (Routing with Constraints and Heterogeneity), to quickly solve large-scale multiple Hamiltonian path problems (mHPP) with traversability constraints, assigning areas to be explored to drones, wheeled robots, and legged robots and determining the order in which those areas are visited.
AFFORD2ACT (Affordance-Guided Keypoints to Act) studies how much manipulation capability can be retained when only sparse 2D keypoints are used as input: an off-the-shelf language-conditioned affordance model locates task-relevant regions, selects up to 15 object keypoints plus 4 end-effector keypoints, compresses the policy input to 19 2D keypoints (38 coordinates), and then uses lightweight gating to decide which keypoints to focus on during execution.
AgiBot World Colosseo introduces the AgiBot World large-scale manipulation platform, containing over 1 million trajectories covering 217 tasks across 5 deployment scenarios, an order-of-magnitude increase in data scale over existing datasets, and uses a standardized collection pipeline with human-in-the-loop verification.
A laboratory study used a commercial software-defined microsegmentation tool to implement dynamic microsegmentation in an Internet-of-Things multi-robot network, with policies adaptively adjusted and automatically deployed based on asset functional and health tags. The demonstration showed that the approach can automatically isolate a designated threat in real time while maintaining normal operation of unaffected nodes, validating the feasibility of microsegmentation for addressing threats and disruptions in IoT networks.
A study proposes introducing a laser pointer into a vision-based portable pipeline for target disambiguation in multi-object scenes, to support grasp selection and preshaping in upper-limb prostheses. The method uses a mixed reality strategy to reuse existing high-quality segmentation masks to generate training images and retrains a deep convolutional neural network. Experimental results show target disambiguation accuracy improves by up to 9% over existing solutions while computational cost remains unchanged.
AOV-GS is an active 3D Gaussian Splatting mapping system that uses frozen foundation models to guide exploration, without requiring manually annotated segmentation masks or closed semantic categories. In its coarse exploration stage, it re-ranks geometric candidate views by the novelty of CLIP mask embeddings computed from SAM instance segmentation, biasing exploration toward semantically informative regions; a lightweight scene language field fitted offline directly aligns the collected CLIP embeddings, enabling open-vocabulary text-query segmentation.
A study evaluated the human activity understanding capabilities of seven locally deployable vision-language models (VLMs) on two person-centric thermal imaging datasets, with diagnostic tasks covering fine-grained action recognition, activity timeline reconstruction, and the ability to withhold answers.
This study examines the degradation of world-model predictions when physical conditions fall outside the training distribution, comparing two causal modeling elements: the inputs available to the predictor and how its mechanism computes their effects. On a synthetic benchmark with shortcut features, a frozen dependency mask derived from paired interventions reduced the out-of-distribution RMSE of a fully connected predictor from 1.769 to 0.697, close to the generator-oracle mask's 0.694.
RoPT is a 6DoF object pose tracking framework for robotic interaction that couples visual perception, temporal kinematics, and global re-localization into a closed loop to counter catastrophic divergence in monocular pipelines under fast motion. Its visual module tracks object masks to handle partial and severe occlusions; a kinematic factor graph regularizes frame-to-frame updates, predicts forward poses, and drives parallel multi-hypothesis verification to avoid local minima; and an active re-localization module gives the system failure awareness and deterministic self-recovery.
PCLA-Bridge is a deployable tool that lets unmodified CARLA Leaderboard agents run from any sensor source to any control destination without an active CARLA server.
The paper proposes AW-0, an agentic world action model framework that combines code-based planning with joint video and action prediction. Within a unified mixture-of-transformers architecture, it simultaneously models three coupled generative streams: code, future video, and robot actions. A code expert generates parameterized skill routines and control logic, while video and action experts generate future visual states and low-level motor commands.
PhysiCalWorld proposes calibrating action-conditioned world models' physics via in-context prompting, without fine-tuning or analytical physics modeling, to adapt to instance-level dynamics.
CW-WAM is a world action model that jointly predicts video, motion, and contact wrench. Inspired by hybrid motion-force control, it uses a unified interaction module to predict time-aligned motion-wrench chunks as control references. It structures motion-wrench dependencies with pair-causal attention and retains past interaction information in a fixed-length wrench memory; the predicted wrench chunk serves as the force reference for a 20 Hz outer-loop admittance controller.
The paper proposes PAPAV, a capability-centric unified framework that distills the recurring functions of multimodal embodied agents (MMEAs) into five: Perceive the current state, Anticipate action effects, Plan a feasible course, Act through an interface or body, and Verify the outcome.
The paper proposes AtomicRL, a framework that uses atomic task structure to improve reinforcement learning post-training of vision-language-action (VLA) policies. Its core components include Semantic-Aware Policy Optimization (SAPO), which uses semantic gating to prevent TD residuals from failed atomic subtasks from propagating across tasks, thereby mitigating credit contamination; and Bottleneck-Adaptive Target Exploration (BATE), which dynamically coordinates task sampling and local exploration incentives based on the mastery level of each subtask.
The study trained 12 policies with the same recipe on a simulated Unitree G1, varying gravity, terrain geometry, and ground contact models (rigid or 3D resistive force theory, 3D RFT), and tested Earth-to-Moon fine-tuning and gravity randomization.
RISE is a perceptive humanoid fall-recovery method that uses task objectives at inference time to steer how the robot stands up. It learns a recovery vocabulary that links intended movements to motor actions through sensory feedback; a visual diffusion planner learns from executions of this vocabulary to predict motion intent as future body states, and task costs steer that prediction toward a desired heading, rise speed, or path, after which motion latents are generated accordingly. Simulations demonstrate diverse recoveries, terrain adaptation, and test-time steering without retraining, with a real-robot demo on the Unitree G1.
The study proposes a unified closed-loop manipulation evaluation framework that assesses frontier AI models' ability to control a robot directly through visual observations, robot proprioception, and structured tool calls on the LIBERO-10 benchmark, without task-specific fine-tuning.
SP-HLM proposes a sparse-pose-driven interface framework that uses five task-critical body poses plus a learned implicit motion prior to represent spatial intent and whole-body coordination in complementary forms. The SP-VLA vision-language-action model first predicts task-relevant sparse poses and finger joint positions from multimodal observations and language instructions via asynchronous action-chunk inference. The SP-MGM motion generation model, trained through two-stage distillation, reconstructs coordinated full-body motion in real time, and the SP-GMT general motion tracker then combines the explicit sparse targets with the implicit motion prior for whole-body control.
TARSphere is a robot-oriented full-stack framework for long-horizon embodied agents operating under continuously changing physical and social environments. It includes TARSphere-Infra, which runs robot policies in physics simulation; TARSphere-Harness, which lets a frozen multimodal large language model make decisions through multi-level executable skills; and TARSphere-Bench, an evaluation benchmark comprising 360 episodes across 30 indoor scenes.
DONIC is the first generalist bimanual dexterous tracker jointly trained across manipulation categories and human motion sources. It is released alongside BiHOI, a multi-view dataset containing 1,575 sequences covering 229 objects, in which reconstructed interactions are retargeted into robot references.
An arXiv paper proposes an object- and task-aware active perception framework that lets a high-degree-of-freedom mobile manipulator use its hand camera to actively inspect uncertain or safety-critical regions during navigation. The framework uses a probabilistic next-best-view (NBV) formulation and is integrated with hierarchical task model predictive control (HTMPC), exploiting kinematic redundancy to adjust viewpoints while preserving the higher priority of base-trajectory tracking. Real-world tests showed that, compared with passive or non-task-aware baselines, the system improved map consistency and reduced task-related map entropy and navigation risk.
HuMiT is a whole-body humanoid teleoperation system whose unified tracking policy requires only a compact reference target—root height, root velocity, and the SE(3) poses of four sparse keypoints at the current frame—yet achieves tracking accuracy comparable to methods that rely on full joint positions, dense body-part transformations, and even future frames.
RangeNav is a LiDAR range-image encoder with only 183,000 parameters, pretrained entirely on synthetic data, that uses anisotropic convolutions and circular padding to handle the extreme aspect ratio and azimuthal seam of range images.
This study proposes a spline-based movement primitive representation that unifies spatial geometry and temporal progression into a common set of continuous variables, jointly estimates a shared Bezier curve, and allows spatial and temporal model complexity to be adapted independently. Simulations validated joint recovery accuracy, refinement from modifications to model complexity and continuity, and improved robustness to variance in the demonstration set. In a robot-assisted patient-turning task, online retiming and synchronization capabilities were evaluated with a physical manikin.
An arXiv paper formalizes zero-shot generalization in physical AI as coverage of the deployment distribution, noting that more samples from the same collector can only slowly shrink the unseen region, while each additional contributor can expand coverage. Based on this, the paper proposes a shared latent space for robot actions that combines multimodal alignment, generative augmentation, and noisy-label learning, and suggests that developers build applications around the standard robot Origmi to provide diverse data. It concludes with a testable hypothesis and evaluation protocol linking real-robot zero-shot success rates to data diversity.
An arXiv paper proposes OMAR, an open-world multi-attribute reasoning pipeline for inferring a shelf's latent organization and ranking placement positions for new objects without user preference instructions. A vision-language model first proposes object labels and attributes such as content and packaging type, which are grounded through segmentation, depth, and multi-view association; an LLM then reasons over an instance-level 3D scene graph to produce scene-consistent placement rankings, followed by geometric and kinematic feasibility checks before execution.
The paper proposes post-hoc kernel-variance heads that estimate epistemic uncertainty from the embeddings of a single policy, without regression targets or gradient updates, and without changing the original policy.
An arXiv paper proposes distinguishing two goals in autonomous system failure analysis, reproduction and avoidance, and formalizes the concept of "necessary causal explanations," which together with existing "sufficient explanations" forms a unified causal analysis framework. Based on actual causality and causal responsibility, the paper presents an efficient approximation algorithm for generating necessary explanations, evaluates its efficiency across multiple environments, and uses counterfactual generation to verify that it outperforms existing methods on failure avoidance.
SpatialDex is a framework that learns 3D-grounded bimanual dexterous manipulation from RGB-only human videos through reconstruction-augmented visuomotor pre-training, learning human behavior together with its spatial context. It reconstructs scene geometry and estimates hand motion from each video, aligns the two at a video-level scale using the hand-size prior of 3D hand estimates, and uses the reconstructed geometry as both a policy input and a prediction target.
CausGround proposes a non-parametric causal interaction grounding framework that distills interaction experience into reusable, action-centered causal knowledge, identifies action preconditions and effects, and organizes them into a factorized world-state representation. At planning time, it aligns task-relevant causal dependencies with current observations, supporting backward reasoning over preconditions and forward reasoning over effects.
AirPlayground is a unified benchmark for training and evaluating aerial navigation policies, built on 94 photorealistic 3D Gaussian Splatting (3DGS) environments spanning indoor, urban outdoor, and vegetation settings, with aligned collision geometry.
FlowVLA separates action generation from physical-completion estimation. A shared π0.5 action expert learns approach, pour, and return, while a target-independent estimator fuses visual, joint, wrist-wrench, and recurrent force-belief histories to estimate transferred mass. It triggers return when this estimate plus predicted residual outflow reaches the requested mass.