AI 119 articles
AI
ActSafeGuard Puts Hard Joint Limits Inside Flow-Matching VLAs
SJTU’s ray-scaling layer hits 100% step safety on π0.5 and Fast-WAM, then keeps a real ALOHA arm on a magnetic-ball track.
AI
FARM Reads Failure Out of a Frozen World Model
AI
FreqFM Conditions VLA Flow Matching on Action Frequency
AI
ReactHuman Freezes the Scene and Asks the Model to Catch or Dodge
AI
2AM Keeps Task Memory in the Agent, Not the Robot Policy
AI
HuRo Paints a Robot Over 630,000 Human Videos
AI
IMLE-VLA Drops the 10-Step Action Head and Hits 55 Hz
AI
MaP-WAM Remembers as a Plan, Not a Growing Video
AI
UniMPA Grounds Predicted Futures in Actions the Robot Has Already Done
AI
GTA-2 Builds Robot Skills From Axes, Not End-to-End Actions
AI
HaWMPO Down-Weights Hallucinated World-Model Rollouts
AI
JEPA Policy Skips Diffusion and Still Beats It on Five Robot Tasks
AI
Show-Harness Lets a VLM Play the Robot Like a Game
AI
PPS Steers a Frozen π0.5 Without Touching Its Weights
AI
FailureSpot Marks the Instant a VLA Starts to Blow It
AI
RoboSPA: Pick the Third-Farthest Bottle, Then Remember What You Hid
AI
AGC-VLN Lets a Drone Draw the Map a Ground Robot Actually Drives
AI
GIFT Trains Robot Policies to Keep Geometry, Affordances, and Goals
AI
GPT-6 Astra Drops a Block in a Bowl 19 Times in 20. The Puzzle Still Stalls
AI
RoboTok Indexes Web Video by 3D Hand Trajectories, Not How the Kitchen Looks
AI
AGIBOT Opens 11,430 Real Trajectories That Include the Failures
AI
Astribot’s SmoothRL Fine-Tunes a VLA While the Robot Is Still Moving
AI
EGR Stops a VLA From Listening to the Wrong Camera
AI
FWBC-VLA Wipes a Board and Opens a Loaded Door Without a Force Sensor
AI
WISE Only Imagines When the Robot Is About to Touch Something
AI
XR-2 Trains on 1,500 Hours of Two-Handed Housework, Then DAgger
AI
Facet-0 Puts Force in the VLA — 82% on Sub-Millimeter PC Assembly
AI
Isaac Teleop Can Now Buzz a MANUS Glove Finger by Finger
AI
Perceptron Opens Isaac 0.5, a 36B Robot Model
AI
Skild S1 Learns a 10-Minute Job From One Video
AI
Anthropic’s Model Hardware Standard Lets Claude Drive Lab Robots
AI
CLAP Trains Video World Models Across Robot Bodies, Including G1
AI
Jetson Orin Nano 2: 78 TOPS, Same Board, Ships 2027
AI
Figure Opens Index: 16 Million Human Videos for Helix
AI
GVLA Stops Treating Every Gripper Like a Parallel Jaw
AI
LD4WAM Learns Robot Actions From 5,000 Hours of Human Video
AI
ViTacPhys Estimates Mass and Stiffness Before It Squeezes
AI
GEN-1.5 Learns a New Robot Task From a 3-to-12-Second Demo
AI
NVIDIA’s SONIC Turns Motion Tracking Into a Humanoid Foundation Model
AI
Illinois Fine-Tunes π0.5 With the Robot’s Own Failed Rollouts
AI
RoboEdit Turns Human Hand Videos Into 14 Million Robot Frames
AI
Hydra-0 Treats Robot Actions as Pixel Motion
AI
τ0-VLA Thinks About the Next Subtask Before It Commits
AI
CoRe Recovers a Frozen VLA by Imagining the Next Move
AI
HAF Splits a Generalist VLA So a Humanoid Can Walk and Work
AI
ViTaR Lets a Frozen VLA Feel Contact Without Rewriting It
AI
ART Lets a VLA Call Night-Vision Instead of Retraining
AI
BICPO-VLA Smooths the Handoff When the Next Chunk Is Late
AI
ReflexVLA Chases Moving Cubes Instead of Frozen Scenes
AI
DaViNCi Makes Street Navigation Continuous — and the Cars Move
AI
DreamX-Phi Predicts What a Bimanual Arm Will See Next
AI
MiDAS Adapts a Generalist Robot Policy From One Demo
AI
H2R-Bench: Pretty Robot Videos Still Fail the Transfer
AI
Seeker Learns Where to Look From the Action Stream
AI
Temporal GRPO Stops Punishing the Good Part of a Failed Rollout
AI
A Linear Probe Reads Task Progress Out of π0.5
AI
Galaxea G0.5 Puts Reasoning and Action in One Token Stream
AI
StellaVLA Adapts From One Demo by Teaching the Why
AI
BooST Moves Franka Skills Onto a UR3 From Five Demos
AI
FACT Trains World-Action Models on Failures Without Copying Bad Moves
AI
Flex-π Predicts RGB, Depth, and DINO Futures From One Checkpoint
AI
Gated VLA-Cache Recomputes Only When the Policy Sounds Unsure
AI
Surgical WAM Turns Unlabeled Endoscopy Video Into Closed-Loop Control
AI
HarnessWAM Adds Planning Memory on Top of World Action Models
AI
JEPA-WAM Couples Latent Transitions With Action Generation
AI
RynnValue Turns Timestamps Into Robot Rewards at 7,000 Hours Scale
AI
SLIM-0.5B Packs Action-Grounded Latents Into a Compact Robot Policy
AI
Vid2WAM Distills Video Diffusion Into a Compact World-Action Student
AI
XPolicyLab Wants One Adapter Per Robot Policy, Not One Per Benchmark
AI
AtlasVLA Remembers the Room With Only a Wrist Camera
AI
Cross-View Consistency Makes Flow VLAs Survive Camera Moves
AI
CrossTracer: One Semantic Path, Residual Fixes per Robot
AI
Dyna-2 Trains a World-Action Model on 1M Hours of Human Video
AI
LifelongCrossNav Keeps a 3D Map Across Floors and Goals
AI
TEMPO Splits VLA RL: Freeze Semantics, Train Actions Fast
AI
Adaptive-WAM: Stop Generating the Future Video Just to Steer
AI
EvoHIL: Human-in-the-Loop RL That Survives a Lighting Change
AI
GORDON Learns Object-Centric Rewards From Action-Free Video
AI
PanoVLA Puts 360° Context Into Mobile Bimanual Manipulation
AI
SpaceVLA Puts User Grasp and Place Anchors Into OpenVLA
AI
Zero-Shot VLM Controllers Often Aren't Looking at the Camera
AI
GUARD Watches Whether a Diffusion VLA Is Actually Looking at the Scene
AI
HiRoC Post-Trains VLAs as Planner Plus Subgoal Executor
AI
LiLa-WAM Trains a World-Action Model on One 24 GB GPU
AI
Mimir Splits World Memory from Task Memory for Long-Horizon Agents
AI
Robust-WAM Keeps Video Pretraining and Adds Semantic Foresight
AI
SpikingNav Navigates With Fewer FLOPs and Better Corruption Survival
AI
DyPES-VLA: Shared Dynamics Priors, Native Actions Per Embodiment
AI
GeniWorld Turns Robot Actions Into Pictures the World Model Can Follow
AI
In-Context VLA: Consume Grounded Language, Don’t Narrate CoT
AI
JoyAI-RA 0.5: Scale Manipulation by Aligning Human Video to Robot Actions
AI
SkillMemo Stores Atomic Skills and Pulls Them Back at Runtime
AI
VLAff Learns Where, How, and Which Way From Ego Video
AI
W²-VLA Predicts How the Wrist Will Move Before It Moves
AI
BridgeVLA++ Adds Spatio-Temporal Memory Without Blowing the Data Budget
AI
SiMDex: Mine 5% of Ego Videos, Beat Random Mixing on Dexterous Tasks
AI
SARF Defends VLA Robots Against Physical Attention-Hijacking Patches
AI
Track4Action Distills 3D World Tracking Into Tracker-Free VLA Policies
AI
HERO Bootstraps Robot Skills From Zero Human Demos
AI
HoloMotion: Horizon’s Open Whole-Body Humanoid Controller Hits 300 FPS
AI
HumanCLAW Asks: Can VLMs Act Through a Body?
AI
Robbyant Open-Sources LingBot-VLA 2.0 Across 20 Robot Morphologies
AI
CLIFT Fine-Tunes Closed Gemini Robotics On-Device into Near-Perfect Humanoid Skills
AI
X Square Robot Opens Its Embodied AI Stack: Data, World Model, Action
AI
BrainCo Demos Thought-Controlled Robots at WAIC 2026
AI
Humanoid’s KinetIQ Ascend Pushes Real-World RL Past Demo Manipulation
AI
LimX COSA 0.5: Whole-Body VLA That Tidies a Room Uncut
AI
RoboParty UFO: Open Unsupervised RL for Humanoid Control
AI
Encord Is Tagging Brain Waves While Pilots Train Robot Arms
AI
Flexion, Niantic Spatial, and NVIDIA Close a Sim2Real Loop for Humanoids
AI
Naver’s ARCBRAIN Flow Aims at No-Code Robot Service Design
AI
Gemini Robotics 2: Whole-Body Control From Feet to Fingertips
AI
General Intuition Raises $320M to Train Agents on Gameplay Action Data
AI
NVIDIA GR00T 1.7 and Isaac Teleop Land in Hugging Face LeRobot
AI
Sanctuary AI Pivots Physical AI Onto Today’s Factory Arms
AI
Generalist’s GEN-1 Learns Across Thousands of Robot Hands
AI
Interlatent Opens a Low-Latency VR Teleop Stack (QUIC + Browser IK)
AI
Enigma Raises $71M to Make Robot Control Feel Obvious
AI