Research

Papers

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

1 papers

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

LevelAny Beginner Intermediate Expert

Oct 1, 2026arXiv

Video world modelsIntermediate

PhysicsLENS: Diagnosing Physical Property Blindness in Video Generation Models

Isaiah Milkey, Som Sagar, Aditya Taparia, Xinyuan Liu, Jiqing Wen, Ransalu Senanayake

This arXiv preprint introduces PhysicsLENS, a benchmark that tests whether robot video generation models follow text-specified physical properties that are not visible in the starting frame. Across four video generators and 439 clips, human raters found 34 of 47 plausible-looking clips ignored the stated latent property, a gap current visual-quality evaluations miss.

World Models

Read the full paper (PDF)

Topic pages

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

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