Skip to content
#Robot models and learning
Oct 27Tue
  1. arXiv cs.RO58

    Can Differentiable Simulators Give Better Policy Gradients? ICLR 2026 Paper Proposes DDCG and IVW-H

    An ICLR 2026 paper re-examines the question of whether differentiable simulators can give better policy gradients, arguing that bias from discontinuous dynamics is not the main obstacle. The authors propose DDCG, a lightweight test that switches estimators in nonsmooth regions; it requires only one hyperparameter and remains reliable with small samples. They also propose IVW-H, a per-step inverse-variance implementation that stabilizes variance in differentiable robot control tasks without explicit discontinuity detection. The paper concludes that estimator switching can improve robustness in controlled experiments, while in real deployment variance control is often more critical.

  2. arXiv cs.RO58

    ActECC: An Action Chunk Temporal Scheduling System for Edge-Cloud Collaborative VLA Inference

    The authors propose ActECC, which addresses the difficulty of determining the valid range of action chunks and scheduling replanning requests in edge-cloud collaborative VLA inference through kinematics-guided usable range assessment and delay-aware replanning scheduling. Experiments show it reduces per-action latency by up to 27.4%, cuts cloud calls by up to 45.1%, and lowers average execution error.

  3. arXiv cs.RO58

    Vision-Language-Action Model Robot Learning Framework for Agricultural Manipulation

    A research team proposed a simulation-to-real robot learning framework for agricultural reaching and grasping, building a MuJoCo-based agricultural environment with diverse tomato plant structures and collecting expert demonstrations via joystick teleoperation. Two pretrained generalist policies, Octo and π0.5, were first adapted with simulated demonstrations, then fine-tuned with real demonstrations on a Franka Research 3 robotic arm. Results showed that π0.5 scored higher than Octo on task performance and degraded less under distribution shifts such as viewpoint and spatial offsets; the study also examined the effect of proprioceptive information on policy performance.

  4. arXiv cs.RO63

    CLARO: Open-vocabulary grasping in cluttered scenes with a VLM agent

    CLARO is a VLM agent that orchestrates skills such as open-vocabulary object detection, segmentation, 6-DoF grasping, and non-prehensile pushing to locate and retrieve a language-specified target in cluttered environments. It records past actions to support long-horizon execution and outputs grasp poses or 3D waypoints, enabling transfer to different robot arms without task-specific training. In a dynamic simulation environment, CLARO achieved an average success rate of 89.1%, a relative improvement of over 16% on the strongest competing method in highly cluttered scenes; on a real robot arm, its success rate was 88.8%.

  5. arXiv cs.RO71

    PlayPolicy: Learning Dexterous Manipulation from Supervised Play

    The authors propose the PlayPolicy framework, in which a robot first learns a play policy from minutes of human play data through behavior cloning, then deploys it autonomously to collect hours of interaction data to train a world model; at inference time, this policy proposes candidate action sequences for model-predictive control (MPC).

  6. arXiv cs.RO55

    Active Perception of Upper-Limb Functional Mobility Through Human-Robot Handovers

    This study formulates upper-limb functional mobility estimation as an active perception problem, parameterizing reductions in healthy joint limits with impairment coefficients and integrating this into a human model used during robot interaction. The authors propose an online active perception pipeline in which the robot actively probes the user through repeated handover requests while progressively updating a Bayesian belief over the impairment coefficients. Simulation studies and a proof of concept on a robotic platform show that the pipeline can recover the impairment coefficients with relatively high accuracy in real human-robot handovers.

  7. arXiv cs.RO60

    CapPro: Adapting VLA models to new manipulation tasks with a single human video

    Researchers propose the CapPro framework, which lets a vision-language-action (VLA) policy adapt to previously unseen manipulation tasks using only a single human demonstration video, without robot demonstrations or fine-tuning for the target task. The framework has two components: a phase-aware human video representation module, which learns task-relevant query representations through visual token selection and phase-caption supervision, and a video-conditioned adaptation module, which injects the compact video and query representations into the action-conditioning context of a pretrained VLA policy.

  8. arXiv cs.RO62

    OneDemo: Learning policies and generating data from a single demonstration

    OneDemo proposes a reinforcement learning framework that edits a single successful demonstration into trajectories suited to new task configurations; the learned editor serves both as a task policy and as a reusable demonstration generator. On 15 Meta-World tasks, it reaches a 92.7% success rate using an average of 53.7k environment interactions; on 9 DexMimicGen tasks, it generates demonstrations with an average success rate of 98.4%, and BC-RNN policies trained on the generated data reach an average success rate of 99.6% at the same data scale.

  9. arXiv cs.RO62

    DiVeR: Decision-Critical Verifier Learning for VLA Test-Time Scaling

    DiVeR estimates decision criticality from the dispersion of sampled action representations and uses it to reweight verifier learning, focusing learning on states where action selection matters most, without requiring step-level annotations or additional environment interaction. In experiments on LIBERO, RoboCasa, and a Franka Research 3 real robot, DiVeR consistently improves task success rates by guiding action selection with a more effective verifier, and verifier inference overhead is negligible.

  10. arXiv cs.RO60

    DGT-Map: Directional Global Traversability Mapping with Multi-Task Learning for Heterogeneous Vehicles

    DGT-Map proposes a self-supervised framework that learns global, directional, and vehicle-conditioned traversability costmaps from RGB-D observations and locomotion signals. It uses a shared multi-task backbone to learn common terrain features across vehicles, while vehicle-specific prediction heads preserve platform-dependent responses. At inference, it outputs a heading-indexed costmap for use by direction-aware planners.

  13. arXiv cs.RO59

    EPCC: Faster-than-Demonstration Planning with Slow Expert Demonstrations and Fast Play

    The authors propose Expert-Play Contouring Control (EPCC), which combines slow expert demonstrations with fast, non-expert play to achieve execution speeds faster than the demonstrations. Expert demonstrations train a latent trajectory generator whose predictions are reparameterized into a time-independent contour of task progress, while play data trains a world model of fast-action outcomes. At deployment, the planner optimizes actions along this contour and penalizes deviation from it.

  14. arXiv cs.RO67

    VLN on the Fly: An Onboard Vision-Language Navigation Stack for Aerial Robots

    The authors propose VLN on the Fly, an onboard vision-language navigation stack that keeps instruction grounding, planning, and control as separately inspectable stages: a quantized VLM grounds the instruction to a coarse image cell, depth lifts it to a 3D goal, a fast B-spline planner generates a feasible trajectory, and a pretrained reinforcement learning policy tracks the trajectory and outputs motor commands across multiple quadrotors.

  15. arXiv cs.RO65

    ForceDelta-VLA: Distilling Force-Conditioned Action Corrections for Contact-Rich Manipulation

    ForceDelta-VLA proposes a correction-distillation framework that constructs an explicit force-correction target from the force-conditioned and force-agnostic predictions of a frozen teacher model, and adds a separate delay-correction target to handle reference-action mismatch. Across nine single-arm and bimanual contact-rich tasks, the system achieves a mean success rate of 82.2%, compared with 54.4% for the ForceVLA baseline and 70.6% for direct execution of the Stage-1 temporal teacher. Compared with ForceVLA, the full system reduces the mean peak contact force of successful trials by about 26% on both platforms.

  16. arXiv cs.RO78

    X-WBC: A Cross-Embodiment Foundation Model for Humanoid Whole-Body Control

    X-WBC is a cross-embodiment humanoid whole-body control foundation framework that separates relatively shared human motion semantics from embodiment-specific physical execution. It uses human-centered command tokens to align full human motion, robot reference motion, and sparse VR observations; a causal Transformer learns reusable temporal structure from mixed multi-robot rollouts, and lightweight robot-specific modules map the shared representation to each robot's proprioception and action space.

    Why it matters: The paper proposes a cross-embodiment humanoid whole-body control framework that decouples shared human motion semantics from each embodiment's execution, and validates joint training on nine simulated embodiments and four real robots.

  17. 雷科技70

    Xingyuan Intelligence releases cross-embodiment data engine Chorus, claims about 40x training data expansion

    On October 27, Xingyuan Intelligence released Chorus Zhongxing, a cross-embodiment data engine for robot pretraining that can convert human manipulation videos captured by smart glasses, chest-worn cameras, wrist cameras, UMI devices, smart headbands, and other devices into pretraining data adapted to different robot embodiments.

  18. arXiv cs.RO47

    Robot Motion as a Keyed Communication Channel: Transmitting Messages via Motion Steganography

    Researchers propose a "motion communication" method that encodes arbitrary short messages (such as an agent's current intent) as noise into the actions of any pre-trained policy, making robot motion detectable via remote vision or motion capture without degrading policy performance. This physical channel does not rely on wireless communication and requires no additional hardware or direct link to the robot. On real robots running at 50 Hz, four robots jointly recovered an 8-bit message at a total rate of 0.8 bits/s, with validation completed in both simulated environments and real-robot deployment.

  19. arXiv cs.RO77

    AgiBot Team Releases GE-Act 2.0 World-Action Model Technical Report

    The AgiBot Research Team released a technical report on GE-Act 2.0, a world-action model whose generative and action components are all trained from scratch on manipulation data. It consists of a control-oriented autoencoder (CoAE), a single-step visual planner (SVP), and an inverse dynamics model (IDM), and is jointly trained with KASO.

    Why it matters: The original text presents the component design and data-scaling curves for pretraining a world-action model from scratch, allowing observation of changes in zero-shot cross-embodiment transfer.

  21. arXiv cs.RO71

    Learning Loco-Manipulation From SMPC Demonstrations With Sparse Offline-to-Online RL

    This work proposes using sample-based model predictive control (SMPC) in simulation as an automated expert to generate large-scale offline datasets, then training an off-policy reinforcement learning agent with purely sparse task rewards, eliminating the need for manual dense reward design. Combined with a low-level dynamic stability controller, the high-level agent's learned policies can exceed the original optimal control teacher on task objectives. The authors completed sim-to-real deployment on an arm-equipped Spot quadruped and a G1 humanoid; the paper is the accepted version for CoRL 2026.

  22. arXiv cs.RO72

    DSMS: Contact-Implicit Multiple Shooting for Dynamic Motion Retargeting

    The study proposes DSMS, a contact-implicit, direct-simulation-based multiple shooting framework that converts kinematically feasible reference motions into dynamically feasible whole-body trajectories. It handles contact, friction, collision, and joint limits within a nonlinear program without presetting a contact schedule. Compared with existing retargeting methods, DSMS accelerates motion-imitation reinforcement learning training and yields policies with high success rates and low tracking error. The authors achieved zero-shot sim-to-real transfer on the Unitree G1, completing walking, jogging, contact-rich crawling, and a 180-degree jump-turn.

  23. arXiv cs.RO72

    SMAT: Jointly training human-exoskeleton co-adaptation policies in simulation and validating on real subjects

    The study proposes Staged Multi-Agent Training (SMAT), a method that jointly trains a musculoskeletal human agent and a bilateral hip exoskeleton control policy in physics simulation to learn an assistance policy that accounts for human-device co-adaptation. The policy was transferred to a physical hip exoskeleton without subject-specific retraining and tested in eight healthy adults; indirect calorimetry showed that active assistance reduced net metabolic rate by 19.7% relative to the passive device (p < 0.001), the hip mechanical positive-power ratio was 0.984, and effective assistance was maintained at walking speeds and terrains beyond the training conditions.

  25. arXiv cs.RO67

    LaWAM: Efficient Dynamics-Aware Robot Policies with Latent World Action Models

    LaWAM proposes a Latent World Action Model that exposes predictive dynamics to robot policies through compact latent visual subgoals rather than reconstructed future video. Its core LaWM trains a latent action model in the latent space of a pretrained vision foundation model and reuses its forward decoder to predict future observation features.

  27. arXiv cs.RO60

    PRISM: Prior-Guided Imagination Sampling in World Model Representations

    The authors propose PRISM, which trains a lightweight action head on features from a frozen latent world model encoder to obtain a goal-conditioned action prior, then uses its mean and predicted uncertainty to guide action search, with the world model evaluating the consequences of actions. Experiments on four simulated control tasks and two real-robot platforms show that the lightweight action head improves planning performance, and that the same fusion interface can accept a diffusion prior over the same frozen features without modification.

  28. arXiv cs.RO61

    MRO-GWM: Action-Conditional and Object-Centric Gaussian Splatting World Model for Rigid Objects

    The authors propose MRO-GWM, an object-centric Gaussian world model for learning action-conditional dynamics of 3D rigid objects. The model represents arbitrary shapes and multi-object scenes with Gaussians in an object-centric frame, and uses a temporal-spatial Transformer to predict rigid body motion from past Gaussians and future actions; training is based on multi-view reconstruction, which requires handling partial observations caused by occlusion.

  29. arXiv cs.RO74

    Crazyflow: A JAX-based GPU-accelerated differentiable drone simulator

    The authors propose Crazyflow, a differentiable drone simulator built on JAX that combines high-fidelity dynamics with GPU acceleration, scaling to millions of environments or swarms of thousands of drones on a single CPU and GPU. Real-world experiments show that it supports analytical-gradient policy learning, achieving high-accuracy tracking without domain randomization, and sampling-based obstacle avoidance at over 500 million steps per second. The authors also demonstrate in-flight reinforcement learning: after tossing a physical drone into the air, training a recovery policy from scratch took only 0.37 seconds, and the drone was caught before landing.

  31. arXiv cs.RO58

    LLM-Guided Future Hypotheses for Horizon-Aware Exploration in Multi-Step Robot Manipulation

    The paper proposes Future-Experience Conditioning (FEC), which uses a latent representation of a short future video as a condition for closed-loop policies, for exploration and policy adaptation in multi-step robot manipulation. Future clips are generated in three stages: an LLM reasoner performs task-ontology reasoning initialized from the current scene state, a robot-free digital twin rolls out the intended object motion, and a mask-free video diffusion model synthesizes a robot-consistent future clip.

  32. arXiv cs.RO67

    WorldArena 2.0 released: expanding the embodied world model benchmark across modality, functionality, and platform

    WorldArena 2.0 has been released, expanding embodied world model evaluation along three dimensions: modality, functionality, and platform. In modality, it extends from vision-only to visuotactile; in functionality, it extends beyond policy evaluation and planning to using world models as interactive reinforcement learning environments for policy optimization; in platform, it extends from simulation-only to multiple simulators and real robot embodiments. The benchmark evaluates perception quality, interaction utility, and cross-platform performance under a unified protocol, and the project page provides code and demos.

  33. arXiv cs.RO72

    RoboEvolve: A Planner-Simulator Co-Evolution Framework for Robotic Manipulation

    RoboEvolve couples a VLM planner and a VGM simulator into a mutually reinforcing co-evolutionary loop, bootstrapped from unlabeled seed images to alleviate manipulation data scarcity. Experiments show it raises the base planner by 30 absolute points and improves simulator success by 48% on average, surpassing fully supervised baselines with only 500 unlabeled seed images and reducing initial reinforcement learning data by 50 times.

  34. arXiv cs.RO60

    BEACON: Cross-Domain Co-Training of Generative Robot Policies via Best-Effort Adaptation

    BEACON proposes a theory-guided cross-domain co-training framework that trains diffusion-based visuomotor policies using abundant source demonstrations and limited target demonstrations, and formulates co-training as a discrepancy-aware importance reweighting problem. The method introduces scalable instance-level discrepancy proxies, stochastic alternating updates of policy and weights, and a multi-source extension, improving robustness and data efficiency over target-only, fixed-ratio co-training, and feature-alignment baselines in sim-to-sim, sim-to-real, and multi-source manipulation settings.

  35. arXiv cs.RO60

    SAS: A Semantic Autonomy Framework for VLM-Integrated Indoor Mobile Robots

    The paper proposes the Semantic Autonomy Stack (SAS), a six-layer integration framework that builds semantic reasoning, mission interface, and operational memory layers on a validated ROS 2/Nav2 foundation. It uses a deterministic-first cascade that escalates unresolved or guarded cases to a pretrained vision-language model (VLM).

  36. arXiv cs.RO62

    Visibility-aware mobile grasping system for dynamic environments

    The authors propose a unified system that simultaneously constrains collision risk along the whole-body trajectory and target visibility during grasping in unknown and dynamic environments, comprising three parts: a velocity-aware gaze policy, a hierarchical subgoal policy, and a receding-horizon whole-body planner. In 400 randomized simulation scenarios and real-world trials on a Fetch mobile manipulator, the system achieved success rates of 70.4% in unknown static environments and 57.9% in dynamic environments, 16.3 and 9.3 percentage points higher than decoupled navigation and manipulation policies, respectively, while significantly reducing collisions.

  37. arXiv cs.RO62

    High-Speed, Scalable Sensor Readout Architecture for Dexterous Hands Based on Shift Registers

    The paper proposes a scalable analog sensor readout architecture based on serial-in parallel-out (SIPO) shift registers, in which sensor modules can be expanded using only three signal lines. The authors integrated 16 joint sensor modules and one four-channel tactile sensor module on a tendon-driven dexterous hand, achieving full scanning of 20 channels at 1 kHz and stable operation up to 1.5 kHz; the maximum slope absolute percentage error of the joint sensors was 0.446%, and on the tactile side an LSTM model performed force estimation and contact-location classification at 1 kHz.

  38. arXiv cs.RO62

    SentryVLA: Uncertainty-Guided Action Selection and State Monitoring for Vision-Language-Action Robotic Control

    SentryVLA is a lightweight runtime wrapper for frozen VLA policies that requires no retraining or changes to the model architecture. At the action level, it uses conformal quantile regression to estimate an upper bound on the deviation between sampled actions and expert behavior, and ranks multiple stochastic candidate actions accordingly; at the execution level, it uses a Mahalanobis-distance anomaly score to monitor abnormal states during closed-loop control. Simulation and real-world experiments show that the method improves action selection and provides useful runtime warning signals.

  39. arXiv cs.RO67

    AffordSim: A Scalable Data Generator and Benchmark for Affordance-Aware Robotic Manipulation

    AffordSim is a scalable data generator and benchmark that integrates open-vocabulary 3D affordance prediction into simulation trajectory generation. Given a natural-language task description, it synthesizes scenes, generates affordance queries, and samples region-constrained grasps, then uses motion planning to filter executable candidates.

  40. arXiv cs.RO60

    PrepTurn: An anticipatory turning reinforcement learning framework for fast sequential quadruped navigation

    PrepTurn is a reinforcement learning framework for anticipatory maneuvering in continuous local navigation. It maps the current and next pose goals directly to joint actions, letting the upcoming goal influence how the robot approaches the current one. The framework uses a persistent-credit reward to encourage completing the entire goal sequence while in motion, and an automatic curriculum that gradually increases goal difficulty to stabilize turning behavior; simulation and onboard Unitree Go2 experiments show high sequential traversal completion rates and reduced stop-and-go motion.