Skip to content
#Research
Oct 23Fri
  1. HuggingFace Daily Papers62

    S^2: Sensorimotor Alignment as an Offline Proxy for Embodied Policy Success

    The study proposes a training-free offline metric, Sensorimotor Alignment (S^2), that measures how well a model's sensory representations align with the geometry of physical motion. The authors report that S^2 correlates positively with downstream robot task success across model families, and that it can be used to screen models before closed-loop evaluation and to identify promising VLM backbones before costly VLA fine-tuning.

  2. HuggingFace Daily Papers75

    DiscoDemo: Using reinforcement learning to generate diverse robot demonstration data to improve imitation learning

    DiscoDemo uses reinforcement learning as a data generator, training a privileged-information latent-conditioned generator that discovers multiple successful solutions through a parallel reverse curriculum and a success-gated diversity objective, then exports them after safety filtering as a static dataset for downstream visuomotor imitation learning. On four manipulation tasks, the method raises downstream success in simulation from 40% for the strongest baseline to 62%, and on a real robot from 39% to 78%. The authors believe that the diversity of successful solutions can improve imitation learning for a pretrained flow-matching VLA, and that it complements corrective data collection.

  4. HuggingFace Daily Papers75

    RynnUnify releases 16B unified embodied model combining visual planning and robot control

    RynnUnify is a 16B unified embodied model that combines embodied understanding with long-horizon visual planning through interleaved video-language reasoning, covering spatial reasoning, physical prediction, and grounding. The authors also built an agentic extension using reinforcement learning and introduced a world-action extension that translates understanding and prediction capabilities into robot control.

  5. Tech Xplore Robotics65

    ETH Zurich uses mouse muscle tissue to enable tiny biohybrid robots to jump, swim, walk and grasp

    A research team at ETH Zurich published a study in Nature Communications in which they used computer simulation to co-design mouse skeletal muscle tissue and a soft hydrogel scaffold as an integrated whole, producing a biological actuator about 1.5 centimeters long with a range of motion up to 11 times that of existing systems with the same amount of muscle.

  6. Kinetix AI · YouTube62

    KAI Halo case study EgoWild2Dex: training bimanual dexterous manipulation with first-person data

    The KAI Halo case study EgoWild2Dex shows how high-quality bimanual dexterous manipulation capabilities can be derived from highly challenging egocentric data. The project page is https://mmlab.hk/egowild2dex.

    Why it matters: The material presents a complete training path from first-person human data to robotic bimanual manipulation, along with success rates; its transferable aspect lies in its approach to data utilization.

  7. HuggingFace Daily Papers58

    FrontierCS 2: A Benchmark for Computer Science R&D Released

    Researchers introduce FrontierCS 2, a benchmark of 48 open-ended tasks. Their approach takes a paper and the code its authors released, deletes the paper's contribution, and uses the remaining code as the human reference baseline. Unlike earlier benchmarks that only give a problem description and grade one program after the run, FrontierCS 2 places agents inside a research environment where they inherit a runnable repository and can submit patches at any time. The evaluator returns only a development-suite score, while the evaluator, the held-out final suite, and the reference remain private.

  8. 简智机器人 · X73

    Genrobot open-sources Gen-HumanWholeBody first-person whole-body manipulation dataset

    Genrobot has released and open-sourced Gen-HumanWholeBody, a first-person whole-body manipulation dataset it says has the industry's highest precision, hosted on Hugging Face. The dataset contains 600 hours of whole-body manipulation data, approximately 3 cm whole-body MPJPE, 16,976 real-world scenes, and 1,281 human skills, and is open for research on whole-body motion learning, human-to-robot transfer, and embodied foundation models.

    Why it matters: The original text provides the dataset's scale, precision, and open-access entry point, which readers can use to assess its usability for whole-body motion learning research.

Oct 22Thu
  3. 北京理工大学 · 微信公众号74

    BIT team publishes meter-scale titanium alloy single-crystal additive manufacturing results in Nature

    A team led by Professor Liu Changmeng at Beijing Institute of Technology has pioneered a single-crystal additive manufacturing method based on unsteady pulsed growth in a micro-melt pool, achieving meter-scale titanium alloy single-crystal additive manufacturing for the first time. The results were published in Nature on October 21 in a paper titled "Additive manufacturing ultralong and architected single crystals," with corresponding authors Liu Changmeng, Cui Yinan of Tsinghua University, and Yinmin Morris Wang of the University of California, Los Angeles.

    Why it matters: The paper presents the mechanism and cross-material validation of seed-crystal-free single-crystal additive manufacturing, which readers can use to assess its impact on the barriers to manufacturing single-crystal components.

  5. Xbotics具身智能实验室 · 微信公众号62

    HKU team proposes YOCO for dexterous-hand teleoperation mocap calibration with few paired poses

    Researchers from the University of Hong Kong, Yisheng Technology, and Shenzhen Hetao College have proposed YOCO, a few-shot session calibration method for 3D keypoints, aimed at correcting bias in low-cost motion-capture gloves used in dexterous-hand teleoperation. The paper has been accepted to CoRL 2026.

  6. Xbotics具身智能实验室 · 微信公众号72

    TaskAnchor lets VLA use execution history to address task-state confusion in long-horizon manipulation tasks

    TaskAnchor connects execution history to VLA's existing language and vision interfaces through a lightweight adapter, aiming to alleviate task-state confusion in long-horizon manipulation. It uses a dual-path design: a normalized task-state coordinate that describes semantic execution stages, and a visual residual based on context-adaptive test-time training (CA-TTT) that preserves historical details.

  7. HuggingFace Daily Papers56

    SERA: A Self-Evolving Robotic Agent Framework for History-Dependent Manipulation

    The study proposes SERA (Self-Evolving Robotic Agent), which organizes a high-level VLM into a robotic agent through an editable harness, with a fixed VLA executor providing low-level control. The harness uses observations and execution feedback to maintain task context, and the self-evolution mechanism proposes and evaluates harness revisions through environment rollouts, with the selected configuration fixed before held-out testing. On RoboMME, SERA autonomously selects a configuration reaching 51.38% overall success rate, versus 44.38% without self-evolution.

  8. 机器之心 · 微信公众号55

    ReGDiff: Exploring Metamaterial Voxel Geometry with Guided Diffusion in a Controlled Latent Space

    A team led by Dawei Zhou at Virginia Tech and researchers at Meta propose ReGDiff in a NeurIPS 2026 paper to generate metamaterial voxel structures that satisfy geometric constraints. The method has two steps: first, an RAS mechanism organizes the autoencoder latent space into a form where good and bad structures are clearly separated and good structures are evenly distributed; then, short-range repulsion (SRR) pushes latent variables away from the near neighbors of known samples during sampling, and it takes effect only at inference time.

  9. HuggingFace Daily Papers47

    PAPAV: A Unified Capability Framework for Multimodal Embodied Agents from Computer Use to Robot Use

    The paper proposes PAPAV, a capability-centric unified framework that characterizes multimodal embodied agents (MMEAs) through five functions: Perceive, Anticipate, Plan, Act, and Verify, and uses it to compare multimodal agents (MMAs) with robotic systems (RSs). An analysis across 64 benchmarks shows that explicit evaluation centers on Act, with only 5 assessing Anticipate and 3 assessing Verify. Physical constraints make errors harder to reverse, and PAPAV provides a common basis for designing reliable physical agents and measuring progress across the five capabilities.

  10. arXiv cs.RO67

    Let It Be Simple: Research on One-Step Action Generation for Vision-Language-Action Models

    The paper finds that VLA action generation differs from text-to-image generation: merely biasing the training timestep distribution toward the noise endpoint yields strong one-step policies on LIBERO, LIBERO-Plus, LIBERO-Pro, and real-world robot tasks. For open-source VLA models with different architectures, directly replacing iterative decoding with a single forward pass often performs comparably to the original multi-step decoding on LIBERO, CALVIN, and SimplerEnv.

  11. arXiv cs.RO56

    TrackDeform3D: A Markerless 3D Keypoint Tracking and Dataset Collection Framework for Deformable Objects Using RGB-D Cameras

    TrackDeform3D is an affordable and autonomous framework for collecting 3D datasets of deformable objects using only RGB-D cameras. It identifies 3D keypoints and robustly tracks their trajectories with motion consistency constraints, producing temporally smooth and geometrically consistent data.

  12. arXiv cs.RO46

    Verifiable Safe Synthesis Method for Stochastic Neural Control Barrier Functions (SNCBFs)

    This study proposes a verifiable safe synthesis method for stochastic neural control barrier functions (SNCBFs), providing a verification-free synthesis framework and a verification-in-the-loop synthesis framework for the smooth activation function and ReLU cases, respectively. The method was validated on three cases: the inverted pendulum, Darboux, and unicycle models.

  14. arXiv cs.RO60

    SSIO: A Self-Supervised Inertial Odometry Method Without Ground-Truth Trajectories

    The paper proposes a self-supervised inertial odometry training scheme called SSIO, in which a myopic teacher independently integrates the IMU within each short window to supervise the student model, an approach called integration self-supervision. Although the student is supervised only by short-window velocity changes, it learns to predict velocity directly and does so more accurately than its integration teacher. Pose graph distillation is then used to further refine the model with a cross-window consistency objective.

  15. arXiv cs.RO55

    PaceVLA: Adaptive Temporal Action Quantization in Vision-Language-Action Models

    PaceVLA proposes an adaptive temporal execution framework for VLAs that uses multi-granularity action decoders on a shared action expert, along with quantizability-guided mode selection and context-conditioned routing, to decide at each inference step whether to adopt coarse-grained execution. Experiments cover three multi-task simulation benchmarks and six real-world tasks, improving execution efficiency while maintaining task success rates across multiple embodiments, including long-horizon and dexterous manipulation.

  16. arXiv cs.RO77

    Learning from Fleet Experience: Collective Experience Learning Across 96 Robots

    The paper proposes the Collective Experience Learning (CEL) paradigm, in which deployed robots pool their own execution experience to update a shared policy that is then returned to the fleet for use. It is implemented through failure-guided experience retrieval fine-tuning (FERT), which uses failure contexts to retrieve relevant segments from successful trajectories and prioritizes their recorded actions for supervised updates, without additional expert demonstrations or corrective actions; a supporting fault-tolerant asynchronous training system (FATS) coordinates robot execution, cloud inference, and policy optimization.

    Why it matters: The paper proposes replacing expert corrections with the deployed robots' own execution experience to iterate on the policy, and reports real-world manipulation results at a scale of 96 robots.

  17. arXiv cs.RO60

    VQ-WM: A Latent Feedback Motion Planning Framework for Robot Manipulation

    The paper proposes Vector-Quantized World Model (VQ-WM), a latent feedback motion planning framework that combines discrete long-horizon search with continuous visual feedback control: a discrete latent state predictor models transitions between vector-quantized latent states, allowing the planner to search for the shortest path over long distances, while a continuous learned controller then reaches the planned states step by step through visual feedback to correct deviations and mitigate compounding errors.

  20. arXiv cs.RO51

    FallGown: A Soft Pneumatic Wearable Device for Fall Prevention

    FallGown is a soft wearable support structure made of pressurized air pouches inside a fabric shell; it can support the full weight of a user's body to prevent falls. When stored it measures 6.9 kg and 0.051 cubic meters, with an average donning time of under 2.5 minutes and doffing time of under 0.5 minutes. Single-user fall prevention testing showed a tradeoff between device size and resistance to tipping; a configuration with a base, 12 air pouches, and a skirt-bottom diameter of 1.4 meters prevented all falls except for one mechanical failure. Onboard pressure sensors can sense the user's weight distribution, offering potential for monitored exercise and adaptive support.

  21. arXiv cs.RO58

    RoboCap: A six-camera head-mounted platform and Grounded API for egocentric robot learning

    The research team introduces RoboCap, a 250-gram head-mounted device with six cameras and dual IMUs for in-the-wild egocentric data collection, along with the Grounded API, a device-agnostic suite of 3D algorithms. The report shows how the hardware, calibration, and 3D algorithms work together to achieve state-of-the-art performance on public benchmarks, covering SLAM across diverse settings and rigs, egocentric depth estimation, and hand tracking when adapted to third-party devices.

  22. arXiv cs.RO60

    Robot Learning with Visual Predicted Force: Achieving Force-Aware Manipulation Through Visual Force Prediction

    The paper proposes achieving force-aware manipulation by visually predicting the deformation of a compliant Fin Ray gripper, without requiring force or tactile sensors at deployment. The method trains two models: first, a visual force estimator is trained on calibration data to annotate demonstrations with force estimates; second, an action-force proposal policy is trained to jointly generate candidate actions and their corresponding forces. At test time, candidate actions are sampled and the action whose predicted force is closest to a target from the demonstrations is executed. The method was evaluated on four tasks: berry picking, empty-can grasping, in-hand reorientation, and plug insertion.

  23. arXiv cs.RO62

    PolyUMI: A visual-tactile-audio multimodal data collection platform for object inference and manipulation

    PolyUMI is an open-source visual-tactile-audio demonstration collection and robot deployment platform. Its lightweight wireless handheld gripper synchronously records wrist-camera, optical tactile, contact-audio, and proprioceptive observations without requiring a tethered workstation. The same sensing finger can be transferred to the robot end effector to keep the sensing geometry consistent between data collection and policy execution.

  24. arXiv cs.RO57

    HarnessPAI: An Evolvable Harness Framework for Physical AI

    HarnessPAI proposes a model- and embodiment-agnostic Harness framework for Physical AI that treats code as the executable, evolvable interface organizing underlying action primitives. The framework separates two timescales: within a single rollout, it executes open-loop at the program level with a fixed program guiding and checking execution; across rollouts, it evolves closed-loop, using execution feedback to revise the program and distill failures into reusable skills.

  25. arXiv cs.RO55

    BAS-VLA: A Task-Semantic Action Calibration Framework for VLA Models

    The paper proposes BAS-VLA, a task-semantic action calibration framework built on top of a frozen base VLA, designed to address two types of failure: policies producing unnecessary action drift when task semantics are unchanged but appearance varies, and policies continuing along the original trajectory when key semantics such as the target object or constraints are altered.

  26. arXiv cs.RO54

    rMuscle: A Real-Time Framework That Uses Robotic Task Similarity to Accelerate VLA Model Inference

    The paper proposes rMuscle, a real-time vision-language-action (VLA) model inference framework inspired by human muscle memory. The authors first analyze embodied workloads and find that when robots repeatedly perform tasks, there is significant similarity in observations, action trajectories, and even internal model states.

  27. arXiv cs.RO51

    Action-Conditioned World Model for Robotic Neck Ultrasound Target Plane Guidance

    This study proposes a local action-conditioned world model framework for probe guidance toward target planes in robotic ultrasound, using a two-stage model-based learning pipeline: first, a latent conditional diffusion world model predicts future ultrasound observations from recent context frames, probe motion, and temporal offset; second, a goal-conditioned temporal Transformer predicts ordered probe actions and is fine-tuned using rewards from the frozen world model.

  28. arXiv cs.RO55

    Hilti-Trimble-Oxford dataset released: a benchmark for 360-degree visual-inertial SLAM and floor plan localization

    The paper releases the Hilti-Trimble-Oxford dataset, collected at an active construction site. It contains 30 visual-inertial sequences covering 10 floors over an eight-month span, and captures real-world challenges such as changing lighting, moving workers, fast motion, and repetitive structures. Ground-truth trajectories were collected with a high-precision LiDAR-inertial SLAM system rigidly attached to the 360 camera. The authors also report results from the accompanying open research challenge: 62 teams entered the SLAM track, while only 22 entered floor plan localization, indicating less attention to the latter.

  30. arXiv cs.RO67

    ReStruct: Guiding Robot Behavior at Inference Time via Physically-Aware Reconfiguration of Task Structure

    ReStruct proposes a method to steer learned robot policies at inference time, enabling them to satisfy user preferences not specified during training without retraining. The method decomposes a visuomotor policy into a high-level state-machine skeleton and a low-level residual controller, and incorporates preferences into the skeleton via a synchronous product to reconfigure the task structure, while the controller remains frozen. Simulation and real-world experiments show that ReStruct can handle a variety of preferences, from object-level specifications to temporal-logic constraints, and after steering it outperforms existing methods in task success rate and preference adherence, leading VLA models by up to 25%.

  31. arXiv cs.RO60

    SafeDojo: Safe Reinforcement Learning for VLA via Interactive World Models

    SafeDojo proposes a model-based safe reinforcement learning framework that improves vision-language-action (VLA) policies through world-model imagination, avoiding costly real-world exploration or reliance on hand-crafted safety functions. The method performs online reinforcement learning on top of an interactive video world model, using a ResNet success classifier to score predicted future frames, paired with a lightweight safety head that predicts per-step contact probabilities from the current latent context and candidate action chunks, then balances task execution and trajectory safety signals through Lagrangian-constrained GRPO.