Skip to content
Only Research#Other
Oct 22Thu
  1. 机器人前瞻 · 微信公众号73

    TeleAI and others propose NEXUS, a cross-terrain perception and motion control foundation, enabling Unitree G1 to perform full-body actions on stairs and slopes

    The China Telecom Artificial Intelligence Research Institute (TeleAI), Gamma Intelligence, Shanghai Jiao Tong University, the University of Hong Kong, and Zhejiang University jointly proposed NEXUS, a cross-terrain perception and motion control foundation for humanoid robots that lets robots autonomously adjust posture and contact based on terrain while preserving motion intent.

  2. arXiv cs.RO51

    AgriVLN: A Vision-and-Language Navigation Benchmark and Method for Agricultural Robots

    This work proposes A2A, a vision-and-language navigation benchmark for agricultural robots, comprising 1,560 episodes across 6 classes of agricultural scenes, using step-by-step instructions and a continuous-action environment. Evaluation found that existing VLN methods perform well on short instructions but poorly on long instructions; to address this, the authors propose an STL module that decomposes complex instructions into lists of simple subtasks and integrates it into the NavGPT backbone, raising SR from 0.33 to 0.47 while NE slightly increases from 2.76 m to 2.91 m. Ablation studies and case analyses are also included.

  3. arXiv cs.RO55

    Distance-Field-Based Reinforcement Learning for Reactive and Robust Obstacle Avoidance in Robotic Arms

    This paper proposes a reinforcement learning framework that integrates the Euclidean Distance Field (EDF) directly into the observation space, providing the agent with continuous geometric cues to achieve stable and responsive obstacle avoidance behavior and simplifying reward design; it effectively avoids obstacles both with and without artificial potential field guidance.

  7. arXiv cs.RO38

    Real-Time Ground Normal Vector Estimation for Spherical Mobile 3D Mapping Using LiDAR-Inertial Odometry

    The study proposes two RANSAC-based ground normal vector estimation methods: single-scan estimation and estimation based on a LiDAR-inertial odometry (LIO) global 3D map, for odometry in a spherical mobile mapping system (SMMS). Evaluated on diverse real-world datasets against terrestrial laser scanner ground truth, the LIO method significantly outperforms the single-scan method in accuracy and robustness. Integrating ground normal vector estimation into the motion model improves pose estimation and supports autonomous navigation in hazardous environments.

  8. Tech Xplore Robotics57

    University of Illinois proposes CARL method to keep reinforcement learning stable for wheeled, flying, and six-legged robots

    Researchers at the Grainger College of Engineering at the University of Illinois Urbana-Champaign combined reinforcement learning with control theory to propose a contraction-aware reinforcement learning method, CARL, which guarantees robot stability while improving tracking performance by up to 15% compared with existing methods.

Oct 21Wed
  2. arXiv cs.RO47

    Real-Time Disturbance-Aware Koopman Model Predictive Control for Quadrotors in Spatially Varying Winds

    The study proposes a disturbance-aware Koopman control family (DA-KCF) framework that preserves an offline-identified nominal model, learns only a disturbance correction online, and embeds it into model predictive control (MPC) to optimize position and yaw references. The reference-dependent KCF is first learned from simulation data and then calibrated with limited hardware data; during deployment the nominal model remains fixed, and a lightweight residual ensemble learns a low-dimensional disturbance correction from causal state-command-state transitions.

Oct 19Mon
  1. Xbotics具身智能实验室 · 微信公众号72

    Xiamen University HumanPlus, CMU, and MIT Launch CLOTHO, Embedding IMUs in Loose Clothing for Eight-Hour Full-Body Motion Capture

    The HumanPlus lab at Xiamen University, together with researchers from MIT and CMU, has launched CLOTHO, a full-body motion capture system that integrates 10 IMUs into loose clothing. The paper was accepted to the SIGGRAPH Asia 2026 journal track and will be published in ACM Transactions on Graphics.

  2. 36Kr Embodied Intelligence66

    AgiBot, NVIDIA, and universities launch WorldOracle world-model evaluation platform

    WorldOracle, a world-model evaluation platform jointly developed by researchers from AgiBot, NVIDIA, and several universities, is now live and open for evaluation submissions. The platform uses 335 real manipulation trajectories, with about 40% of interaction segments containing failed or suboptimal operations, testing model prediction against real execution outcomes. Testing of 7 world models found that while some models can reproduce robotic-arm motions relatively accurately, their predictions of object states still show considerable deviation, and performance drops markedly on long-horizon prediction.

  3. 雷科技78

    WorldOracle world-model evaluation platform launched, using real-robot testing to fact-check the industry

    On October 16, WorldOracle, an action-driven world-model evaluation platform jointly completed by researchers from AgiBot, NVIDIA, Princeton University, the University of Texas at Austin, and the University of Toronto, was officially released, opening evaluation submissions to developers.

    Why it matters: The evaluation compares world-model predictions frame by frame against real execution records, revealing a common problem in which robotic arm motion is replicated but object states are systematically hallucinated, providing concrete reference points for judging model usability.

  5. arXiv cs.RO38

    Multi-Agent Reinforcement Learning-Based UAV Path Planning for Obstacle Avoidance in Stochastic Environments

    This study proposes a centralized training with decentralized execution method based on multi-agent reinforcement learning for UAV obstacle-avoidance path planning in stochastic environments, improved from the model predictive control (MPC) idea. Each agent communicates only with the centralized planner, plans feasible paths through extended observations based on a distance-weighted mean field approach, and uses multi-step value convergence to improve training efficiency. Comparative, ablation, and real-machine experiments verified the method's effectiveness and generalization performance.

Oct 18Sun
Oct 17Sat
  1. 机器之心 · 微信公众号70

    HKU's EgoWild2Dex trains dual-arm robot on 538.9 hours of "extremely dirty" human data, with under 1 hour of real-robot data

    The University of Hong Kong's MMLab team proposes EgoWild2Dex, which directly trains a dual-arm dexterous robot on 538.9 hours of unscripted human first-person manipulation data, using under 1 hour of real-robot data per downstream task; across three long-horizon tasks—unpacking a parcel, fixing ornaments with a hot glue gun, and making a cold drink—the average success rate is 96.7%.

  2. Tech Xplore Robotics47

    Team develops embodied robotic arm EmArm with human-like tactile sensing

    A team at the Hong Kong University of Science and Technology developed the embodied robotic arm EmArm, embedding a soft optical tactile skin into a rigid skeleton to achieve whole-arm tactile sensing and adaptive manipulation, with tactile sensing fidelity reaching 97.9% under five environmental disturbances. Its tactile skin covers over 71,000 square millimeters with a spatial resolution of 0.085 millimeters, can recognize five touch behaviors and provide emotional feedback, with recognition accuracy up to 99.8%. The research was published in Nature Sensors.

  4. 3D视觉工坊 · 微信公众号41

    Monocular depth-assisted joint optimization calibration of LiDAR-camera extrinsics, moving beyond grayscale mutual information, with superior accuracy and robustness under complex lighting and materials

    A two-stage joint optimization calibration method for LiDAR-camera extrinsics assisted by monocular depth is proposed, replacing traditional grayscale-intensity mutual information with depth constraints to improve calibration accuracy and robustness under complex lighting and material conditions. In the coarse calibration stage, a composite calibration board combining ArUco markers and a circular-hole array is used, and an SVD-optimized ICP algorithm solves for initial extrinsic values analytically without requiring preset initial values; in the fine calibration stage, the DepthPro monocular depth estimation network is introduced to construct a joint optimization objective function combining geometric constraints and normalized mutual information distance (NID).

Oct 16Fri
  1. PaperWeekly · 微信公众号54

    Beihang KDD 2026 paper proposes MoE-Pointer, using Seq2Seq reinforcement learning to jointly optimize rider-drone relay delivery

    Professor Wang Jingyuan's team at Beihang University, together with the School of Management at Shanghai University, published the paper MoE-Pointer at KDD 2026, defining the Dynamic Multi-Stage Pickup and Delivery Problem (DM-PDP) and using a sequence-to-sequence generation framework to achieve end-to-end reinforcement learning joint optimization of a three-stage rider-drone delivery chain.

Oct 15Thu
  1. HuggingFace Daily Papers52

    OneWM unified causal training framework validates mutual gains between world models and action learning

    The paper proposes a unified causal training framework, OneWM, in which actions serve simultaneously as conditions for world prediction and targets for policy learning within a shared video-action architecture. The authors evaluate policy learning and world modeling on LIBERO, LIBERO-Plus, and RoboCasa365, supplemented by real-world policy experiments: learning the visual consequences of specified actions improves policy performance and cross-scene generalization, while action learning under the same architecture improves the fidelity of multi-view world prediction. Positive results were obtained for mutual gains in both directions.

  2. 3D视觉工坊 · 微信公众号43

    Xidian Open-Sources AdaAnchor4D: Monocular 4D Reconstruction for UAVs with Adaptive Anchors Reaching SOTA

    Xidian University and Sun Yat-sen University propose AdaAnchor4D, an adaptive anchor deformation framework for 4D reconstruction of dynamic scenes from monocular UAV footage. It reaches SOTA on PSNR, SSIM, and other metrics across the UAV-Arc4D, VisDrone, and UAVDT datasets while maintaining real-time rendering. Built on 3D Gaussian Splatting, the framework uses three modules—ACFA anchor-conditioned feature aggregation, DLGD decoupled local geometric deformation, and DACW density-adaptive coordinate warping—to eliminate ghosting artifacts and preserve dynamic details.

Oct 14Wed
  1. Tech Xplore Robotics43

    Can Security Robots Be Trusted? University of Michigan Study Reveals How Fairness Affects Trust in Robot Authority

    Xin Ye, a doctoral student at the University of Michigan School of Information, and Professor Lionel Robert published a study in the Proceedings of the Human Factors and Ergonomics Society Annual Meeting. In an experiment with 100 participants involving a parking lot security robot, they found that interactional fairness and distributive fairness significantly increased people's trust in the authority of security robots.

Oct 8Thu
  1. Tech Xplore Robotics48

    Feather star-inspired underwater soft robot achieves 3D maneuverability with only two actuators

    Researchers at North Carolina State University have developed an underwater soft robot inspired by feather stars that achieves three-dimensional maneuverability—up and down, forward and backward, and rotation about an axis—using only two actuators, whereas such motion typically requires at least six. The robot uses four elastic monostable wings and completes hovering, translation, and steering through combinations of a jellyfish mode, a fish mode, and a rotor mode. The research has demonstrated the robot carrying a camera to explore underwater spaces and cooperating with multiple robots to lift objects; a fully wireless version is still in development, and the paper was published in Science Advances.

Oct 7Wed
  2. Tech Xplore Robotics62

    EPFL releases FiberMotor, a fiber motor that uses sliding fibers to drive flexible robots without gears

    Researchers at EPFL's Soft Transducers Laboratory have introduced FiberMotor, a thread-like motor that converts electrical energy directly into motion; the work was published in Advanced Materials. FiberMotor is 1–3 mm in diameter and consists of two concentric hollow fibers. A slightly thicker-than-a-hair insulated copper wire electrode is densely wound around the outside of the fibers; when voltage is applied, electrostatic force makes the inner fiber slide inside the outer fiber like a telescope, so no gears or transmission are needed and it can be driven in reverse.

Oct 6Tue
Oct 2Fri
  1. Tech Xplore Robotics49

    SUTD ALBATROSS: a bioinspired sensing platform that spins down like a maple seed and sails autonomously after splashdown

    A team at the Singapore University of Technology and Design (SUTD) developed the ALBATROSS bioinspired platform, which, when released from 150 meters, descends stably by passive autorotation on a rigid wing-sail, splashes down with low impact and self-rights, then sails autonomously between set waypoints.

Oct 1Thu
  1. Tech Xplore Robotics47

    Toyohashi University of Technology experiment reveals: robots approaching from behind are perceived as faster

    A team at Toyohashi University of Technology conducted five experiments using real autonomous mobile robots (AMRs) and VR; 92% of participants perceived a robot approaching from behind as faster than one approaching from the front at speeds of 0.5–1.0 m/s. A simulation incorporating this perceptual bias into a navigation model showed that, when overtaking pedestrians from behind, the robot began avoiding 0.57 s earlier than with a conventional model and increased passing distance by up to 0.40 m. The results were published in IEEE Robotics and Automation Letters.

  2. Tech Xplore Robotics38

    Can AI agents "feel surprised"? Mutual Information Surprise framework helps autonomous systems respond to the unexpected

    A research team from the Georgia Tech H. Milton Stewart School of Industrial and Systems Engineering (ISyE) published a paper in INFORMS Journal on Data Science proposing a new framework called Mutual Information Surprise, which helps AI agents, robots, and autonomous systems identify "meaningful surprises" and judge whether they need to stop and reassess.

  3. Tech Xplore Robotics63

    NUS team achieves multi-segment bending in soft robotic arm using continuous artificial muscle

    A team from the National University of Singapore's College of Design and Engineering published research in Science Advances using wire-bonded junctions (WBJ) inspired by octopus neuromuscular junctions, enabling a single shape-memory alloy spring artificial muscle to heat and contract in selected segments, achieving multi-segment controllable bending along a continuous arm without needing an independent actuator at each position.

Sep 29Tue
  1. Tech Xplore Robotics60

    MIT team builds paper-thin, two-finned robot powered by muscle cells that can swim through underwater mazes

    An MIT team published in Advanced Functional Materials a paper-thin, two-finned swimming robot powered by a single layer of living muscle cells that can follow light to swim through a simple underwater maze. The robot's skeleton is a GelMA gel film about the size of a stick of gum and roughly half a millimeter thick, stamped with square recessed grooves on its underside; muscle cells on its surface are genetically engineered to twitch in response to light, and researchers steer direction and speed by alternately illuminating the two fins, with the robot able to swim about 4 times its body length in one minute at its fastest.

Sep 24Thu
  1. Chelsea Finn57

    Real-Time EXPO-FT lets π0.5 handle latency and tackle dynamic tasks

    Real-Time EXPO-FT enables π0.5 to tackle challenging dynamic tasks such as balancing a ball and striking a ball into a goal; the paper page is https://pd-perry.github.io/real-time-expo-ft/. Chelsea Finn explains that the approach handles latency in reinforcement learning by having the base VLA use older images while a small editing policy runs in real time, and states that it significantly outperforms RTC across comparisons.

Sep 23Wed
Jul 27Mon
Jun 10Wed
  1. Sergey Levine37

    A New Trick for RL Training of Flow Matching Policies

    Diffusion (or flow) makes for excellent policies, but training them with RL is notoriously hard: BPTT is unstable, and RL over diffusion blows up the horizon. In our new paper, we show how to optimize a flow matching actor with a "weird little trick"—using the identity matrix to "approximate" the Jacobian of the flow denoising process. 👇