RoboScript learns manipulation tasks in context from one demonstration
Authors report 83.4% mean success on all 50 RoboTwin 2.0 tasks without gradient updates, over 20 points above the strongest trained policy.
Stories per day · 30 days
4 RSS
Relatedrobotics13reinforcement-learning4vision-language-action3benchmark2imitation-learning2simulation2world-models2active-perception1assembly1atomicrl1denoising1embodied-ai1fleet-learning1funding1hierarchical-memory1hindsight-relabeling1in-context-learning1manufacturing1
56 stories, newest first · one card per story
Stories tagged #manipulation
Pixel-UMI pipeline for robot whole-body manipulation
New method learns environment-aware whole-body policies from human demonstrations achieving 87% task progress.
HeLM Improves Non-Markovian Manipulation in VLA Models
HeLM achieves 12.8 percentage point intra-episode success gain on 10 non-Markovian tasks with 56 variants.
ActiveAlign uses motion and touch to close Real2Sim2Real gaps
On a Unitree G1, the method supports button presses and slider moves at millimeter scale.
Quality-controllable robot policies use iterative hindsight relabeling
Speed control raises conveyor throughput, and stability control improves success while carrying wine glasses.
Echo denoises stop and failure signals inside vision-language action policies
Echo learns task status without frame-level progress labels and uses those signals to condition later action denoising.
AtomicRL lifts long-horizon VLA robot performance over PPO
Reported gains are 3.6–4.0% on LIBERO-LONG, up to 9.5% on CALVIN, and 21.3% on real robots.
Eka Robotics raises $150 million at a $1.45 billion valuation
The startup’s vision-force-action model and force-sensing gripper are aimed at dexterous object grasping.
RoboValue benchmark tests robotic value models on sim and real tasks
Errors remain in instruction grounding, progress tracking, and recovery even when outcome scores look strong.
VLDA learns compact deltas as robotic action interfaces
Implemented in LoongForge, it targets long-horizon consistency without predicting full future states.
Collective experience learning improves robot policies from fleet failures
Held-out robots reach 93.3–99.2% success on five tasks; a 96-robot phone-packaging run rises from 79.2% to 98.3%.
LIBERO-Industry benchmarks robot policies under factory process constraints
Five generalist policies, including vision-language-action models, degrade as industrial constraints are combined.
AlphaNIST
Researchers train one robot policy to complete all twenty NIST assembly-board tasks using success-guided sampling.
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control
Success Guided Sampling focuses mega-scale robot RL on the policy frontier, solving tasks prior methods miss.
FoldBack: Self-Correcting Masked Generative Policy for Long-Horizon Garment Folding
FoldBack lets a folding robot undo and retry failed grasps without retraining.
TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning
TouchScale's 500 hours of vision and touch raise robot task success from 22.5% to 57.5%.
QF3: Fast Flow RL with Filtered Q-Gradients
QF3 trains robot flow policies off-policy, learning humanoid locomotion from scratch about ten times faster than FPO++.
DepthWorld: 3D World Model for Robot Manipulation
DepthWorld jointly predicts multi-view RGB and metric depth for robot manipulation from calibrated 3D data.
Execution-Aligned Progressive Noise for Consistent Asynchronous Replanning in Generative Robot Policies
EAPN aligns shared noise with executed motion so generative robot policies replan without inconsistent mode switches.
When to Switch: Reliable Action-Chunk Extension for Vision-Language-Action Models
RACE predicts subskill transition timing so VLA models can execute longer action chunks with fewer pauses.
EyeRobot 2.0: Active Gaze for Precise Manipulation without Wrist Cameras
EyeRobot 2.0 achieves wrist-camera-level bimanual manipulation using only active gaze from a single stereo camera, outperforming passive stereo by 40% in real trials.
Skill2Real: Agentic Skill Learning for Zero-Shot Sim-to-Real Robot Manipulation
Skill2Real learns sim-to-real robot manipulation skills via a Proposer-Verifier-Governor loop, achieving 78.75% task completion on real robots without fine-tuning.
World Action Learning via Interaction-Centric Spectral Latent Guidance
WING transfers interaction knowledge from egocentric human video to robot policies via spectral latent guidance.
EyeRobot 2.0: Active Gaze for Precise Manipulation without Wrist Cameras
EyeRobot 2.0 matches wrist-camera manipulation using active gaze and only one stereo camera.