MEKAWIRE
Research

Papers

New robotics and physical-AI papers, with what each one means for the industry.

276 papers
TopicAllManipulation · 27VLA · 22Perception · 12Robotics · 11Navigation · 10Humanoid · 9Learning · 8Safety · 5World Action Models · 4Simulation · 3Aerial manipulation · 2Dexterous manipulation · 2Sim-to-Real Manipulation · 2Social Navigation · 2Vision-Language Navigation · 2World models · 2World-action model · 23D Assembly · 13DGS dataset · 15G Simulation · 1AMR Operations · 1Active vision manipulation · 1Adaptive MPC · 1Adaptive action horizons · 1Adaptive intervention · 1Aerial · 1Aerial Manipulation · 1Aerial navigation · 1Agricultural Robotics · 1Articulation Estimation · 1Autonomous Driving · 1Autonomous Exploration · 1Battery-aware RL · 1Belief-space planning · 1Bimanual Manipulation · 1Byzantine localization · 1Canal inspection · 1Co-speech Humanoid · 1Compositional Manipulation · 1Construction navigation · 1Contact-Rich Manipulation · 1Contact-rich Manipulation · 1Continuum sensing · 1Curriculum RL · 1Dexterous Manipulation · 1Distributed Manipulation · 1Distribution Shift · 1Driving Simulation · 1Efficient VIO · 1Egocentric data · 1Egocentric video · 1Embodied Agents · 1Embodied policy · 1Empowerment MPC · 1Endoscopic navigation · 1Endovascular robotics · 1Event-camera VO · 1Exoskeletons · 1Failure recovery · 1Federated 3D Perception · 1Fisheye detection · 1Floorplan SLAM · 1Geometry-aware manipulation · 1Grasping · 1Grounding · 1Humanoid Control · 1Humanoid RL · 1Humanoid Tool Use · 1Humanoid control · 1Humanoid loco-manipulation · 1Humanoid safety · 1Imitation Learning · 1JEPA World Models · 1Lab automation · 1Laboratory Manipulation · 1Latent World Models · 1Legged Manipulation · 1LiDAR Security · 1Lifelong RL · 1Locomotion · 1MAPF · 1MAV guidance · 1Manipulation LfD · 1Manipulation Safety · 1Marine simulation · 1Memory-centric VLA · 1Miniature Locomotion · 1Mining Perception · 1Mobile manipulation · 1Model-based RL · 1Monocular SSC · 1Motion Planning · 1Multi-Agent RL · 1Multi-Robot Exploration · 1Multi-Robot Manipulation · 1Multi-UAV Exploration · 1Multi-UAV State Estimation · 1Multi-agent RL · 1Multi-agent motion planning · 1Multi-camera odometry · 1Multi-fidelity RL · 1Multi-robot manipulation · 1Multi-skill Manipulation · 1Multi-view perception · 1Multimodal simulation · 1Navigation & Mapping · 1Object Pose Estimation · 1Object Search · 1Panoramic 3D Detection · 1Panoramic segmentation · 1Pedestrian tracking · 1PersonNav · 1Planning · 1Policy · 1Policy Optimization · 1Preference Learning · 1Quadruped locomotion · 1Real-time VLA · 1Real-to-sim · 1Representation alignment · 1Robot-Use Agents · 1Runtime Verification · 1Safe Generative Planning · 1Safe RL-MPC · 1Safe VLA · 1Safety monitoring · 1Safety-critical control · 1Sim-to-real Navigation · 1Sim-to-real manipulation · 1Simulation geometry · 1Skill Discovery · 1Social navigation · 1Soft Pneumatic Actuators · 1Soft manipulation · 1Sonar SLAM · 1Streaming 3D reconstruction · 1Submovement decomposition · 1Surgical robotics · 1Tactile · 1Tactile Exploration · 1Tactile VLA · 1Temporal manipulation · 1Terrain traversability · 1Text-to-3D Policy · 1Tool Co-Design · 1Trajectory Correction · 1Traversability · 1UAV Navigation · 1UAV Swarms · 1UAV traffic monitoring · 1Underwater Robotics · 1Underwater active mapping · 1Underwater perception · 1Unified world-action model · 1VLA Conditioning · 1VLA Defense · 1VLA Robustness · 1VLA evaluation · 1VLA manipulation · 1VLA robustness · 1Video world models · 1Visual Generalization · 1Visual Grounding · 1Visual Reward Models · 1Visuomotor Fusion · 1Visuotactile manipulation · 1World Models · 1World-Action Models · 1
LevelAnyBeginnerIntermediateExpert
arXiv
Object Pose EstimationIntermediate

RYOPO: Bringing End-to-End Category-Level Object Pose Estimation into Real Time

Hakjin Lee, Junghoon Seo, Jaehoon Sim

RYOPO is an end-to-end RGB-D model that jointly detects, segments, and estimates 9-DoF category-level object poses without CAD shape priors or an external instance segmentor. On NOCS REAL275 it reaches 52.5 AP at 5°5cm in joint evaluation and 51.1 AP in all-object evaluation, while running at 31.8 FPS on an RTX A6000. This is a preprint; the speed-accuracy balance matters for real-time manipulation where crop-based pipelines scale poorly with object count.

Read the full paper (PDF)
arXiv
Laboratory ManipulationIntermediate

RoboChemGym: A Protocol-Driven Generative Simulation Framework for Long-Horizon Chemical Manipulation

Chenxi Li, Haiyuan Wan, Rui Li, Jingyuan Li, Sha Zhang, Bohan Feng, Jianbao Cao, Zhangrui Zhao, Di Hu, Wangmeng Zuo, Shixiang Tang, Minting Pan, Dongzhan Zhou

Preprint. The authors built RoboChemGym, a simulation framework that turns written chemistry protocols into long-horizon robot manipulation scenes and expert trajectories, with iterative action/scene refinement and domain randomization. In simulation, generation succeeds on 89.2% of 3-step, 73.4% of 7-step, and 52.3% of 11-step clean protocol tasks; adding 1,000 synthetic trajectories to 10 real demonstrations raised real-robot policy success by about 22–24 points over real-only training. This matters as a scalable data source for autonomous lab robotics.

Read the full paper (PDF)
arXiv
VLAIntermediate

MixVLA: Adaptive Mixing of Non-Invariant Information for Generalizable Vision-Language-Action Models

Pingrui Zhang, Yu Zhang, Pengyuan Wu, Bin Wang, Haoming Song, Xianqiang Gao, ZhaxiZhuoma, Zhigang Wang, Dong Wang, Bin Zhao, Xuelong Li

This preprint introduces MixVLA, a model-agnostic training strategy for vision-language-action (VLA) policies that separates invariant and non-invariant latent representations, then adaptively mixes the non-invariant branch to regularize environment-specific variation. On LIBERO-Plus it posts a 76.2% average zero-shot success rate versus 69.6% for OpenVLA-OFT, and on RoboTwin it lifts π0.5 clean-to-randomized success from 46.0% to 62.4%. Because it needs no extra out-of-distribution data or architecture changes, it is a practical robustness upgrade for existing VLA stacks.

VLA
Read the full paper (PDF)
arXiv
Vision-Language NavigationIntermediate

GeoScaffold: Learning Compact Geometric Latents via Reconstruction for Efficient Vision-Language Navigation

Yixuan Jiang, Wentong Li, An Liu, Zihao Xin, Fulin Tang, Cong Leng, Yang Gao, Jian Cheng

This preprint presents GeoScaffold, a training-only geometric supervision framework for streaming vision-language navigation. A 4B policy fine-tuned with eight geometry query tokens reaches 64.8% success rate on R2R-CE and 59.6% on RxR-CE while using only RGB at inference, and runs at 81 ms per frame on an RTX 4090. The approach avoids resident depth sensors or 3D encoders, making spatially aware navigation cheaper to deploy on edge robots.

Read the full paper (PDF)
arXiv
Curriculum RLIntermediate

Bidirectional Voronoi-biased Exploration Curriculum for Reinforcement Learning

Juri Pfammatter, Kaixian Qu, Clemens Schwarke, Victor Klemm, Marco Hutter

This preprint introduces BVER, a bidirectional curriculum that grows start states from the goal and goals from the initial state distribution, biased toward unexplored task space and steered together RRT-style. On quadruped box climbing it reaches 95% success on a 0.4 m box in about 65% fewer iterations than the best reference-free baseline and is the only reference-free method tested to climb a 0.7 m box. It matters because it removes the need for demonstrations or hand-designed rewards in sparse-reward robot learning.

RL
Read the full paper (PDF)
arXiv
Sim-to-real manipulationIntermediate

SARI: Phase-Split Sim-Real Co-Training for Contact-Rich Manipulation

Xingxin He, Yuxuan Jiang, Haonan Zhang, Chuhan Cui, Kaile Li, Zhongxing Zheng, Caihao Xu, Ziqi Wang

This preprint introduces SARI, a phase-split co-training framework that post-trains a vision-language-action policy on simulated free-space approaches and real contact interactions for contact-rich manipulation. On a lamp task, it reduces real demonstration time by 34.3% and reaches 27.5% success at unseen object placements, where full-task simulation, real-only, and full-task co-training baselines score 0%. The result shows how to allocate scarce real contact data while preserving spatial generalization.

VLA
Read the full paper (PDF)
arXiv
Egocentric videoExpert

World Action Learning via Interaction-Centric Spectral Latent Guidance

Zhiming Liu, Yikun Miao, Ying Chen, Hongrui Yin, Fangqi Zhu, Xiaoyi Pang, Quanxin Shou, Zhengyang Yan, Haodong Wang, Song Guo

This preprint introduces WING, a framework that learns interaction-centric latent actions from egocentric human videos and transfers them to robot policies through low-frequency spectral guidance. In simulation it reaches 99.20% average success on LIBERO, 93.80% on RoboTwin 2.0, and 57.7% on RoboCasa–GR1, and on real-world bimanual tasks it achieves 75.0% average success in the standard setting. It matters because it offers a scalable path to use abundant egocentric video instead of costly robot teleoperation for robot learning.

Read the full paper (PDF)
arXiv
Visual GeneralizationExpert

VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning

Mingyu Park, Samyeul Noh, Hyun Myung, Donghwan Lee

This preprint introduces VIGOR, a model-based reinforcement learning framework that adds latent-space consistency to TD-MPC2 for zero-shot generalization to unseen visual distractions. On six DeepMind Control Suite tasks and four Robosuite tasks, it improves mean zero-shot returns by 3.4% over the second-best baseline on DMC and by 43.6% on Robosuite while matching TD-MPC2's sample efficiency. The work matters because it addresses a two-level compounding failure unique to recursive latent dynamics, which is critical for deploying sample-efficient MBRL in visually varying environments.

RL
Read the full paper (PDF)
arXiv
Mobile manipulationExpert

MobiAgent: Dual-Loop Recursive Policy Self-Improvement for Long-Horizon Mobile Manipulation

Chenzhi Liu, Yue Zhang, Jiehong Lin, Jianan Wang, Bo Wang, Zhongrui Wang, Xiaojuan Qi

This preprint introduces MobiAgent, a dual-loop framework for long-horizon mobile manipulation. It decouples high-level VLM planning from low-level flow-matching skills and adds an automated lifelong learning loop from deployment data. The abstract reports no quantitative results, but the approach matters because it targets compounding errors and interference between locomotion and arm control.

VLA
Read the full paper (PDF)
arXiv
Fisheye detectionIntermediate

FUSEye: Training-Light Fisheye Detection with Overlapping Views and Zero-Initialized Adapters

Wenya Su, Kai Luo, Di Wen, Ruiping Liu, Yufan Chen, Junwei Zheng, Kunyu Peng, Kailun Yang

This preprint introduces FUSEye, a training-light method that adapts a frozen COCO-pretrained YOLO26-x detector to fisheye camera images using overlapping views, zero-initialized residual adapters, and learned cross-view fusion. On WoodScape, it lifts mAP50 from 14.80 to 26.61, reaching 84.3% of full fine-tuning accuracy while training only about 227k added parameters. This matters because it lets robots adapt off-the-shelf detectors to wide-angle cameras without expensive full retraining or calibration.

Read the full paper (PDF)
arXiv
Humanoid ControlExpert

KungfuAthleteBot: learning high-dynamic humanoid motion from video with unified robust recovery

Zhongxiang Lei, Lulu Cao, Xuyang Wang, Tianyi Qian, Jinyan Liu, Xuesong Li

KungfuAthleteBot is a preprint pipeline that turns monocular video of elite martial artists into high-dynamic motions for a Unitree G1 humanoid by repairing trajectory physics, using physics-driven low-kinetic-energy initialization, and training one unified policy for tracking plus fall recovery. On real hardware the robot tracks a five-minute Tai Chi sequence and recovers from arbitrary falls in about 0.7 s, reported as the fastest for a unified policy. This matters because it shows video can replace expensive motion capture for extreme humanoid skills when reconstruction, missing actuation, and recovery are addressed together.

Humanoid
Read the full paper (PDF)
arXiv
Object SearchIntermediate

LOCUS: Landmark-Oriented Container Discrimination Using Spatial Graphs

Taylor Bergeron, Shibani Senthilbabu, Kevin Leahy

This preprint presents LOCUS, a graph neural network that updates CLIP embeddings with spatial context from a scene graph to rank same-class containers when searching for occluded objects. In iTHOR simulation, it reduces expected containers opened per search to 0.257 of all containers, a 24% improvement over TidyBot, and transfers zero-shot to RoboCasa and a Stretch 3 real-kitchen pilot. It matters because real robots must choose among visually identical drawers or cabinets, which existing class-level methods cannot do.

Read the full paper (PDF)
arXiv
Multi-view perceptionExpert

Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis

Keerthi Kaashyap, Dennis Anthony, Akshay Krishnan, Nhi Ngoc Nguyen, Jeremy Collins, James Hays, Shreyas Kousik, Animesh Garg

This preprint introduces SNAP, a self-supervised encoder-decoder transformer that learns transferable geometric scene representations from novel-view synthesis without dense 3D labels. The headline result is that SNAP matches or beats self-supervised baselines on five geometry tasks and approaches heavily geometry-supervised models such as VGGT, while staying robust to camera shifts in robot manipulation. This matters because it shows posed RGB video may replace expensive 3D annotation for spatial foundation models.

Read the full paper (PDF)
arXiv
Construction navigationIntermediate

CORNAV: Construction-Aware Reasoning for Robot Navigation on Active Worksites

Parastoo Ali Pour, Deepak Prakash Kumar, Tommy Zhou, Pramod Khargonekar, Mohammad Abdullah Al Faruque

This preprint presents CORNAV, a navigation system that lets robots operate on active construction sites by combining 2D CAD blueprints, project schedules, open-vocabulary 3D scene graphs, and LLM safety checks. In ablation tests, blueprint grounding raised task success from 13.0% to 72.2% and schedule awareness eliminated hard-zone violations, while safety validation cut hazardous-zone entry from 100% to 0% for mislabeled hazards. The approach matters because it avoids the need for BIM models and lets robots follow site safety constraints that change over time.

Read the full paper (PDF)
arXiv
Dexterous manipulationExpert

PointWAM: 3D World Action Modeling for Dexterous Robotic Manipulation

Chunghyun Park, Beomjun Kim, Seungcheol Park, Heeseung Kwon, Yashu Shukla, Seunghoon Sim, Jinwoo Shin, Minsu Cho

PointWAM is a preprint describing a 3D world action model for dexterous robotic manipulation. It represents the scene and hands as 3D point trajectories in a shared space-time frame, and is pretrained on 1.15 million human demonstration episodes before fine-tuning on robot data. On ten DexJoCo tasks it reaches 69.0% average success, 11.7 points above GR00T N1.6, and on a real humanoid it beats robot-pretrained VLAs by up to 25 points on square-peg insertion.

Dexterous Manipulation
Read the full paper (PDF)
arXiv
Dexterous ManipulationIntermediate

DexJoCo-X: Benchmarking Action Representations for Multi-Hand Dexterous Manipulation

Xiangwei Jiang, Yao Mu, Lixin Duan, Wen Li

This preprint introduces DexJoCo-X, a simulation benchmark and toolkit for comparing action representations across seven dexterous hands and six manipulation tasks using 2,100 balanced demonstrations. Under a single joint policy, function-aligned action slots reach 47.7% mean success over all 42 hand-task pairs, versus 47.0% for native joint coordinates and 33.1% for learned cross-hand latents. The work matters because it isolates representation choices under matched data and training, helping teams build scalable cross-embodiment dexterous policies.

Dexterous Manipulation
Read the full paper (PDF)
arXiv
Simulation geometryExpert

I2CD: Direct Image-to-Convex Decomposition for Simulation-Ready Collision Geometry

Qian Wang, Liam Merz Hoffmeister, Brian Scassellati, Daniel Rakita

This preprint presents I2CD, a method that turns a single RGB image into simulation-ready convex collision geometry by freezing a pretrained Hunyuan3D-2 image-to-3D model and training a small 38M-parameter head. On 227 held-out objects it achieves the best volumetric IoU (65.7%) among eight reconstruction-plus-decomposition baselines and runs 6–37x faster end-to-end (about 0.5 seconds per image). In physical xArm7 pick-and-place, it reduces the time to planner-ready scene geometry from 328 seconds to 11 seconds with 85/100 vs 90/100 success.

Read the full paper (PDF)
arXiv
Lab automationIntermediate

A Passive AI System for Verifying Physical State on Automated Liquid Handlers

Junqiong Joanne Qiu, Zeckria Kamrany, Ananya Anand, Emma Vidal, Yunjia Johanna Bai, Nathan S Chen, Sean Son, Varada Abhyankar, Taryn Jakub, Eleazar Eskin

Preprint. The authors built a passive, external vision system called Labware Setup Checker that verifies deck preparation on an Opentrons Flex liquid handler without modifying the robot's hardware or firmware. It correctly identified a completed plate-and-base assembly in all 29 pre-motion frames and matched the protocol reference in 95.1% of 348 exact-match labware classifications. This matters because it adds non-intrusive setup verification that could reduce manual deck errors in lab automation.

Read the full paper (PDF)
arXiv
Bimanual ManipulationIntermediate

EyeRobot 2.0: Active Gaze for Precise Manipulation without Wrist Cameras

Kush Hari, Justin Kerr, Nidhya Shivakumar, Samarth Mahapatra, Carmelo Sferrazza, Jiahui Lei, Jitendra Malik, C. Karen Liu, Ken Goldberg, Angjoo Kanazawa

This preprint presents EyeRobot 2.0, a bimanual manipulation framework that uses a single fixed stereo camera with active gaze: two eye viewpoints swivel to a 3D fixation point, and the system feeds foveated multi-resolution crops to a behavior-cloning gripper policy. In 7 real and 6 simulated tasks, it lifts real static-stereo success from 27% to about 66.5% (roughly +40 percentage points), matches wrist-camera policies with a 69% vs 64% success rate when wrists are unobstructed, and beats them 48% vs 22% when grasped objects occlude the wrists. This matters because removing wrist cameras can simplify end-effectors and keep vision usable during tool use, clutter, and occlusions.

Read the full paper (PDF)
arXiv
JEPA World ModelsExpert

AVL-JEPA: Preventing Causal Dynamics Information Collapse In Joint Embedding Predictive Architecture World Models

Yikang Qiao, Ling Zhang, Ziying Song, Duan Huang

This arXiv preprint introduces AVL-JEPA, a training method that stops joint-embedding world models from discarding action-conditioned physical dynamics while preserving visual representations. Across four simulated robot control tasks, it reaches 81.5% success on PushT and 88.7% on TwoRoom under Gaussian input noise, versus 15.3% and 65.7% for the baseline, while maintaining competitive clean-environment performance. The work matters for robot planners that must stay reliable when camera or state observations are noisy.

World Models
Read the full paper (PDF)

Topic pages

Manipulation · 27VLA · 21Perception · 12General robotics · 11Navigation · 10Humanoid · 9Robot learning · 8Safety · 5World Action Models · 5Simulation · 3Aerial manipulation · 3Dexterous manipulation · 3Sim-to-Real Manipulation · 3Social Navigation · 3

Papers come from arXiv robotics feeds. Where our summary is not written yet, you see the opening lines of the abstract.