Skip to content
#Research
Nov 2Mon
  1. arXiv cs.RO80

    TacClaw-Infinity: A Large-Scale Vision-Tactile-Language-Action Dataset for Dexterous Manipulation

    The research team released TacClaw-Infinity, a large-scale structured Vision-Tactile-Language-Action (VTLA) dataset for contact-rich manipulation, spanning over 10,000 hours, 1 million episodes, and 1 billion frames across more than 1,000 task types.

    Why it matters: This dataset provides a collection interface and quality control pipeline for ten-thousand-hour-scale tactile manipulation data, and its cross-embodiment transfer validation approach can serve as a reference.

  3. arXiv cs.RO52

    Detect and Suppress: A Mechanistic Defense Against Adversarial Patches on VLA Models

    The authors use a sparse autoencoder (SAE) to analyze the internal representations of a VLA model, identify a feature whose activation strongly correlates with the presence of adversarial patches, and suppress that feature at inference time only when a linear probe detects an attack, thereby improving robustness without fine-tuning the VLA. In adversarial patch attack tests on LIBERO-10, conditional intervention improved the success rate under intermittent attacks, whereas continuously applying the same intervention substantially degraded policy performance.

  4. arXiv cs.RO75

    D4orm diffusion-denoising framework extended into a family of multi-robot trajectory planners

    Based on the dynamics-aware diffusion-denoising framework D4orm, this paper builds a family of multi-robot trajectory planners for different operational requirements, including a decoupled planner, an online receding-horizon planner with feedback control, and a distributed planner. Evaluations with differential-drive and holonomic robots in 2D and 3D environments show that D4orm-based methods find high-quality solutions faster and more reliably than sampling-based optimization methods such as MPPI and a learning-based diffusion model method.

  5. arXiv cs.RO58

    CDIP: Compressed Delayed-Information Projection for Underwater Navigation Under Acoustic Measurement Latency

    The paper proposes compressed delayed-information projection (CDIP), which handles delayed-arrival acoustic position measurements in a 15-dimensional error-state Kalman filter (ESKF), retaining source-associated IMU snapshots and the current-to-snapshot cross-covariances, so delayed corrections can be projected without repeating the filter from the start.

  6. arXiv cs.RO60

    Study: Learning and Transferring Closed-Loop Robot Software

    The research treats complete closed-loop robot policy implementations as reusable execution experience: a coding agent generates policy code from a few successful demonstrations and iteratively improves it using simulation feedback. Across four source tasks in RoboCasa, iterative optimization raised the success rate from 28.3% with initially generated code to 64.2% with validation-selected code, and the learning approach was validated on a physical Unitree G1.

  7. arXiv cs.RO62

    Extending the Franka Emika Panda's Life: A ROS 2 Software Stack with a Reliable Position Interface

    This paper releases an open-source ROS 2 software stack that restores ROS 2 support for the Franka Emika Panda and resolves the long-standing unreliability of its external position control interface. The authors' analysis attributes vibrations and protective stops to external control loop timing and sampling jitter rather than to limitations of the robot itself, and on that basis they propose an asynchronous hardware interface that decouples real-time communication from the ROS 2 control loop, a rate-matching mechanism for slower command sources, and a position-domain reference generation strategy.

  8. arXiv cs.RO60

    UniCross: A Unified Cross-Skill Dexterous Manipulation Synthesis Framework

    UniCross proposes a unified framework that models four dexterous manipulation skills—grasping, relocation, in-hand rotation, and in-hand translation—as the same relational-target formulation, and distills a single policy that does not depend on skill identifiers. The policy performs well on each individual skill, generalizes to unseen objects, and supports long-horizon skill chaining and execution of relational targets unseen during training; the framework also applies to different hand morphologies.

  9. arXiv cs.RO67

    DR-LfD: Decomposing and Reorganizing Demonstrations with Atomic Skills and Visuomotor Policies for Bimanual Manipulation Planning

    The authors propose DR-LfD, a framework that integrates visuomotor policies into a TAMP-gated decision-making system, decomposes human demonstrations into atomic skills based on contact relationships, and then reproduces and reorganizes them using visuomotor policies or object-centric primitives. This method shifts the demonstration data requirement from growing exponentially with the skill sequence to growing with the number of skill types, and on two DexMimicGen tasks it outperforms a diffusion imitation policy trained on 1,000 demonstrations while using only 100 demonstrations. The paper has been accepted by T-RO.

  10. arXiv cs.RO58

    Using a Robotic Arm as an Attitude Actuator: Propellant-Free Attitude Maneuver Planning for Free-Floating Spacecraft

    This research uses manipulator motion for propellant-free attitude control of free-floating spacecraft, modeling it as a trajectory optimization problem with joint and collision-avoidance constraints and solving it with an interior-point method. Across a range of spacecraft-manipulator systems, the method computed rest-to-rest slew trajectories of up to 90° and detumble trajectories with a 40% reduction in angular velocity. The authors also directly compare the manipulator's control authority with that of reaction wheel arrays via momentum and torque envelopes, noting that manipulators could serve as a redundant or even primary attitude control system, with particular promise for in-space assembly and manufacturing involving the capture of high relative-mass payloads.

  12. arXiv cs.RO72

    VEGA: Training Navigation VLAs with In-the-Wild Egocentric Video and Geometric Trajectory Supervision

    VEGA proposes a method for training navigation Vision-Language-Action (VLA) models from unlabeled egocentric navigation videos. It reconstructs local scene geometry from monocular video, samples text, image, and spatial waypoint goals, and generates obstacle-avoidance trajectories, then uses these trajectories to train a flow-matching VLA navigation policy; geometry is used only during training.

  13. arXiv cs.RO65

    GES: Accelerating Morphology-Control Co-Design for Buoyancy-Assisted Robots with Gaussian Evolutionary Specialists

    This study proposes Gaussian Evolutionary Specialists (GES), a reinforcement learning-based co-design framework. By maintaining a population of specialist policies whose design-space territories evolve according to their control performance, it mitigates behavioral diversity collapse, in which a single shared policy converges to the same control strategy across different morphologies.

  14. arXiv cs.RO62

    NDPP-Grasp: Non-Differentiable Physical Plausibility Constraints Guide Task-Oriented Dexterous Grasp Generation

    NDPP-Grasp proposes injecting physical plausibility constraints directly into the denoising process of a task-aligned grasp diffusion model, so that guidance can be applied throughout generation even when the constraints are non-differentiable. Existing diffusion methods first train a grasp diffusion model for task alignment and then apply physical plausibility correction after generation, leaving the generation trajectory itself unguided by physical constraints. The work was accepted to NeurIPS 2026.

  15. arXiv cs.RO69

    TacO: A Benchmark Evaluation of Tactile Sensors for Object Manipulation

    The authors propose the TacO framework to systematically evaluate tactile sensors based on manipulation policy performance, covering four modalities: visual, acoustic, magnetic, and resistive. They train separate policies for three tasks—pick-and-place with unknown mass, object reorientation, and plug insertion—and analyze how sensor properties such as spatial resolution, shear sensing, tactile representation, and material friction affect task performance. The results show that the value of tactile information depends on the sensor modality, material properties, and specific manipulation task, rather than being universally beneficial; the sensors, code, data, and hardware designs will be made public on the project website.

  16. arXiv cs.RO62

    AlphaAdj: A Robot Navigation Framework That Adaptively Tunes Control Barrier Functions with Vision-Language Models

    The authors propose AlphaAdj, a vision-to-control navigation framework that uses first-person RGB observations and a vision-language model (VLM) to adjust the conservativeness of control barrier functions (CBFs) according to visual semantic context. The VLM outputs a scalar contextual caution signal that dynamically modulates the CBF parameter alpha, while a geometry- and speed-dependent cap bounds how aggressive the behavior can be. To cope with VLM latency, inference and the control loop run asynchronously, and the semantic signal is taken directly from token-level output probabilities rather than from full autoregressive text generation.

  17. arXiv cs.RO56

    Specification-aware action distribution optimization framework for robotics foundation models

    The paper proposes a specification-aware action distribution optimization framework that enforces a class of Signal Temporal Logic (STL) constraints at execution time without modifying the parameters of a pretrained robotics foundation model. At each decision step, the method reasons over the remaining horizon via forward dynamics propagation to compute a minimally modified action distribution that satisfies a hard STL feasibility constraint, and it is validated in simulation with state-of-the-art robotics foundation models across multiple environments and complex specifications.

  18. arXiv cs.RO62

    FlowTouch: A View-Invariant Visuo-Tactile Prediction Model

    FlowTouch is a view-invariant visuo-tactile prediction model that uses an object's local 3D mesh to encode information for predicting tactile patterns, enabling tactile feedback even before the object is touched. The method combines scene reconstruction with a Flow Matching-based image generation model. Results show that it narrows the sim-to-real gap and generalizes to new sensor instances, and the generated tactile images can also be used for downstream grasp stability prediction. Code, dataset, and videos are publicly available.

  19. arXiv cs.RO70

    π-StepNFT: A Step-wise Negative-aware Online Reinforcement Learning Framework for Streaming VLAs

    The authors propose π-StepNFT, an online reinforcement learning framework for streaming VLAs that treats transitions in SDE sampling as supervision units and uses variance-normalized transition errors to distinguish positive and negative velocity branches, without computing the final-action likelihood.

  20. arXiv cs.RO72

    CHOP: Counterfactual human preference labels improve obstacle avoidance in visuomotor navigation policies

    The authors propose CHOP, which fine-tunes visuomotor navigation policies with counterfactual human preference labels to align them with human intuition about safety and obstacle avoidance. On the SCAND dataset, multiple advanced models fine-tuned with CHOP reduced near-collision events by an average of 49.7%, decreased deviation from human-preferred trajectories by 45.0%, and increased average obstacle clearance by 19.8%.

  21. arXiv cs.RO71

    WildOS: An Open-Vocabulary Object Search System for the Wild

    WildOS is a unified system for long-range, open-vocabulary object search that combines safe geometric exploration with semantic visual reasoning for complex outdoor environments without prior maps and with limited depth sensing. The system builds a sparse navigation graph to maintain spatial memory and uses a foundation-model-based vision module, ExploRFM, to score the graph's frontier nodes, simultaneously predicting traversability, visual frontiers, and object similarity to enable real-time onboard semantic navigation.

  23. arXiv cs.RO62

    CausalWAM: Learning the Causal Sources of Actions in World Action Models

    The authors propose CWAM (Causal WAM), which adds sparse routing paths to the action head of a world action model to separate the latent variables that truly drive actions from the nuisance variables that scene prediction depends on, while keeping the world modeling objective unchanged. The paper proves that under a sparsity constraint, the latent variables used to generate actions are an invertible function of the action-relevant latent variables alone, and the true structure is identifiable up to a permutation of the latent variables, without requiring sufficient temporal dependence or assumptions about how unobserved states evolve. Synthetic data and real-robot experiments validate the theory and show improved closed-loop success rate efficiency.

  24. arXiv cs.RO60

    Sensorium: An open-source multimodal data collection and curation library for embodied surgical AI

    The authors released Sensorium, an open-source Python library built on ROS and ROS2 middleware that records, processes, annotates, and curates multimodal surgical data from configurable robot and sensor streams, covering operator-side acquisition, configurable processing, semantic annotation, and episodic dataset generation. The library was evaluated on 1516 recordings of phantom, ex vivo, and in vivo porcine experiments involving multiple surgical tasks and sensing configurations, with preliminary usability feedback reported.

  25. arXiv cs.RO62

    PGGT-Loc: Feed-Forward Camera Localization with Primitive-Grounded Geometry Transformer

    The authors propose PGGT (Primitive-Grounded Geometry Transformer), a network that takes a single query image and a set of planar map primitives as input and directly predicts a 6-DoF camera pose, enabling feed-forward camera localization. The method sets new state-of-the-art localization performance on both dense surface maps and minimal layout models. Code and models are publicly available, and the paper was accepted to NeurIPS 2026.

  26. arXiv cs.RO60

    DrawSplat: Atlanta World-Guided Implicit Gaussian SLAM for Indoor Reconstruction

    DrawSplat proposes an Atlanta World-guided implicit Gaussian SLAM framework for indoor reconstruction. The method first estimates camera poses, sparse geometry, and gravity direction through joint point-line tracking, then infers structural planes from tracked lines using the shared gravity component of the Atlanta World prior and converts them into differentiable regularization terms on Gaussian centers and normals. The framework also combines an anchor-based implicit neural Gaussian representation with adaptive structural sampling, achieving robust tracking, high-fidelity rendering, and geometrically consistent reconstruction across multiple datasets.

  27. arXiv cs.RO61

    FOCUS: A First-Order Control Method for Observation-Only Imitation Learning Supervised by Optimal Transport

    FOCUS combines local optimal transport alignment with short-horizon differentiable actor-critic learning for observation-only imitation learning, requiring neither temporally aligned expert trajectories nor expert-aligned initialization. The method matches agent rollout windows to multiple sampled expert segments using a soft best-of-K objective, with the transport cost providing a differentiable imitation reward and a learned critic bootstrapping value beyond the rollout horizon.

  29. arXiv cs.RO67

    MotorSense releases first large-scale paired egocentric video and EMG dataset

    MotorSense is the first large-scale paired egocentric video and surface electromyography (EMG) dataset, containing synchronized video, bimanual EMG, IMU, and audio across 100 hours of unscripted manipulation. The research uses this data for cross-modal pre-training between video and EMG to learn an action organization shared across visual and motor spaces, and transfers it to tasks including action recognition, 3D hand reconstruction, and 3D contact reconstruction. Across four action-related visual tasks and eight video datasets, this sensorimotor-aligned representation outperforms conventional video self-supervised methods.

  31. arXiv cs.RO69

    cuPIQP: A Differentiable Batched QP GPU Solver for Robot Learning and Control

    cuPIQP is a GPU batched differentiable quadratic programming (QP) solver based on a proximal interior-point method, with unified support for dense, sparse, and multistage problems. In benchmarks with up to 16,384 QPs per batch, it solves up to 18.8 times faster than Moreau and 4.26 times faster than eight-core PIQP; in an end-to-end learning task, training is accelerated by 4.6 times over QPAX and 7.8 times over Moreau.

  32. arXiv cs.RO34

    Robot Navigation and Canine Welfare: A Pilot Study Surfacing Challenges for Cross-Species Legibility and Proxy Validity

    A pilot study in a real home environment evaluated a "curve-out" navigation maneuver for canine welfare: a lateral arcing pass-by trajectory triggered at the dog's alert distance. Contrary to the hypothesis, the maneuver elicited a stronger stress response than a straight-line baseline. The authors argue that the maneuver's discontinuous execution, with the robot stopping and reorienting multiple times to approximate the arc, undermined behavioral legibility, which itself became a stressor. The study thus raises the legibility of robot behavior to non-humans and proxy stimulus validity as structural challenges for animal-machine interaction (AMI).

  34. arXiv cs.RO40

    How a team that cannot communicate learns its own state from an opponent's reactions

    A game theory and multi-agent study proposes that a team unable to communicate can learn information no single member can see by reading the opponent's reaction to the group as a whole (a tell); one tell can substitute for messaging among all members. In a solved Gaussian model, the team obtains the truth by committing not to use all of the information, and retains the hidden cost the opponent would have paid; an exact test can determine when this conclusion holds in finite games. The study also analyzes how trust is allocated, noting that two interchangeable members responding in different ways can substitute for a single member's private random source.

  35. arXiv cs.RO65

    ID-WM: A Hierarchical World Model Decoupling State Reachability from Local Control

    The authors propose ID-WM (Implicit Dynamics World Models), a hierarchical world model in which a high-level model samples feasible transitions on the state manifold and a low-level world model acts as an inverse dynamics solver to ground those transitions into precise actions. On three navigation and manipulation test suites, it improves over existing policies and diffusion planners by 38.5%, and by 135% over single-layer world model MPC in complex unseen long-horizon configurations.

  37. arXiv cs.RO71

    ROOF: A Learning Method for Humanoid Fall Recovery in Overhead-Constrained Environments

    ROOF proposes a reinforcement learning method with contact-supported geometric escape guidance that enables proprioception-only humanoid robots to first create clearance and then complete the rise under overhead obstacles. In simulation, the recovery success rate was 88.5% under plates that can only be lifted, compared with 18.9% without geometric guidance; it was 98.6% under free plates with randomized size, mass, and placement. After deployment on a Unitree G1, the policy successfully recovered in 95 of 150 attempts across seven real-world obstacle scenes, including furniture and multi-object arrangements not seen in training.

  39. arXiv cs.RO45

    Hybrid Path Planning Framework for Assistive Wheelchair Navigation in Dynamic Pedestrian Environments

    A study proposes a hybrid path planning framework for multi-goal navigation of assistive wheelchairs, using the Traveling Salesman Problem (TSP) for global task ordering combined with the Potential Field (PF) method and heuristic local planners such as RRT and A* to achieve collision-free navigation, and develops two algorithms, TSP-PF-RRT and TSP-PF-A*. In Monte Carlo simulations of crowded and super-crowded scenarios, the framework achieves efficient multi-goal navigation in dynamic human environments containing heterogeneous static and dynamic obstacles, with TSP-PF-RRT producing fewer or zero collisions and shorter travel distances.

  40. arXiv cs.RO58

    Proof-Carrying Evolution: Language Models Design Formally Verifiable Controllers

    The study proposes Proof-Carrying Evolution (PCE), in which a language model provides both a controller and its certificate during search; only candidates that pass certificate verification can be returned, and the deployed artifact is the control law together with its certificate. The authors compare four language models with classical search and synthesis methods on speed-limited docking, maneuvering-target pursuit, and SO(3) attitude control. Certificates for linear problems are verified using exact rational arithmetic; on SO(3), floating-point verification is followed by exact re-verification of each reported champion. The best run of each method also receives a Lean 4 proof.