Skip to content

#具身智能

0 today

Jun 6Saturday

Synced · WeChat

Daxiao Robotics and NTU Release PhysX-Omni for Simulation-Ready Physical 3D Generation

PhysX-Omni models rigid, deformable, and articulated objects in one simulation-ready 3D generation framework, while PhysXVerse contains over 8.7K physical 3D assets across more than 2.9K categories.

Why it matters: Daxiao Robotics and NTU's PhysX-Omni generates 3D objects with physics properties that run directly in simulators, backed by 8,700+ assets in PhysXVerse.

Jun 5Friday

Synced · WeChat

MetaFine proposes a diagnostic meta-evaluation framework for fine-grained robot manipulation

Southeast University and Peking University researchers introduced MetaFine, a diagnostic meta-evaluation framework that tests fine-grained robot manipulation across understanding, perception, and behavior, and the article says traditional binary success metrics can overestimate fine-manipulation capability by up to 70%.

Why it matters: Southeast University and Peking University researchers propose MetaFine, a meta-evaluation framework that diagnoses robot fine manipulation through understanding, perception and action, and finds success-rate-only evaluation overstates ability by up to 70%.

Jun 4Thursday

QbitAI · WeChat

CVPR 2026: NVIDIA, Tesla, and Waymo hear Xpeng present physical AI

Xpeng presented its world-model stack at CVPR 2026, covering X-World, X-Foresight, and X-Cache; the article says X-Cache cuts about 70% of repeated computation, the second-generation VLA used over 4 trillion training tokens, and the in-car stack reduced inference latency to 80 ms.

Why it matters: XPeng's CVPR 2026 world-model stack cuts redundant compute by about 70% with X-Cache and trains its second-gen VLA on over 4 trillion tokens.

Jun 3Wednesday

NVIDIA Blog

NVIDIA Research Presents Grasping, Autonomous Driving and Agent Training Work at CVPR

NVIDIA Research presented three physical AI papers at CVPR: GraspGen-X was trained on 2 billion simulated grasps, LCDrive cuts reasoning tokens by about half versus text-based reasoning, and NitroGen trains embodied agents across more than 1,000 games and 40,000 hours of interaction.

Why it matters: NVIDIA published three CVPR papers, including LCDrive, which has a driving model output trajectories directly and cuts inference tokens by about half to lower latency.

Synced · WeChat

RSS 2026: Ant Lingbo Proposes Autoregressive Causal World Model for Robot Manipulation with 50 Demos

Ant Lingbo and HKUST introduced LingBot-VA, an autoregressive video-action world model that unifies visual dynamics prediction and action inference, and the paper reports fine-tuning with 50 real-world demonstrations per task plus 92.0% and 91.1% success on RoboTwin 2.0 Easy and Hard settings.

Why it matters: Ant Group's Lingbo and HKUST fold frame prediction and action decisions into one autoregressive model, fine-tuned on just 50 real demos per task.

QbitAI · WeChat

Daxiao Robot and NTU Release PhysX-Omni for Unified Physical 3D Generation

Daxiao Robot and NTU introduced PhysX-Omni, a unified simulation-ready physical 3D generation framework for rigid, deformable, and articulated objects, with PhysXVerse covering 8.7K assets across 2.9K categories and PhysX-Bench evaluating six dimensions including geometry, scale, material, affordance, kinematics, and description.

Why it matters: One pipeline generates rigid, soft and articulated 3D assets, so teams skip separate processing and manual mesh fixes.

Jun 2Tuesday

Synced · WeChat

Turing Award Winner Sutton’s New Paper Argues AI Should Move Toward Enactive Cognition

Banafsheh Rafiee and Richard S. Sutton propose an enactive cognition framework for AI, naming four pillars: experience, perception-action inseparability, autonomy, and embodiment.

Why it matters: Turing Award winner Sutton and Rafiee propose a generative cognition framework that treats intelligence as emerging from experience, perception and action, offering a route to AI's next step that differs from data-driven scaling.

May 30Saturday

Synced · WeChat

NVIDIA and Tsinghua Team's Gamma-World Tops Hugging Face Daily Chart

NVIDIA, Tsinghua, University of Toronto, and Vector Institute released Gamma-World, a multi-agent world model using simplex-based positional encoding and hub tokens to cut interaction cost from quadratic to linear, with 8-player latency dropping from 17.6 ms to 4.5 ms.

Why it matters: NVIDIA and Tsinghua researchers propose the multi-agent world model γ-World, using simplex positional encoding and hub tokens to cut inter-agent interaction cost from quadratic to linear.

May 28Thursday

NVIDIA Blog

NVIDIA Research Advances Robotics From Simulation to the Real World

NVIDIA Research presented 8 ICRA papers on sim-to-real robotics: ScheduleStream delivered a 3x speedup for multi-arm planning, COMPASS reached about 80% success across 20 real-world navigation trials, and Grasp-MPC achieved about 75% real-robot grasping success.

Why it matters: NVIDIA's COMPASS navigates real robots after simulation-only training, hitting roughly 80% success across 20 trials, 4.5x a pure imitation baseline.

Synced · WeChat

ICML 2026: AutoMoT reaches SOTA on Bench2Drive and nuScenes

NTU AutoMan Lab, Harvard, and Xiaomi Auto proposed AutoMoT, a unified VLA driving model using a 4B Qwen3-VL Understanding Expert and a 1.6B Action Expert with asynchronous inference, reaching 89.42 DS and 74.09% SR on Bench2Drive with AutoMoT+.

Why it matters: AutoMoT splits a 4B vision-language model and a 1.6B action expert into asynchronous inference, taking first place on both Bench2Drive and nuScenes, a split-model idea for end-to-end driving.

May 27Wednesday

Synced · WeChat

From Foundation Models to Physical AI, Samsung Moves Into the Core LLM Race

Samsung disclosed three AI efforts—Meki, M2RL, and LiveClawBench—covering a memory-based edge architecture, multi-domain reinforcement learning, and Physical AI evaluation; the article also says Samsung has purchased tens of thousands of GPUs for AI infrastructure, but does not provide model size, training budget, or deployment timelines.

Why it matters: Samsung published three studies, Meki, M2RL and LiveClawBench, covering on-device memory, multi-domain reinforcement learning and physical AI evaluation, showing where it is placing bets on devices and robotics.

May 22Friday

Synced · WeChat

CVPR 2026 | HiF-VLA: A Motion-Centric World Action Model

Westlake University and collaborators introduced HiF-VLA, a motion-centric VLA framework that extracts compact Motion vectors with codecs such as H.264 and uses a joint expert to predict future visual motion and generate action sequences, reporting 31.4GB peak memory and 117.7ms latency under the cited history-window setting.

Why it matters: A Westlake University team proposes HiF-VLA, replacing per-frame visual encoding with H.264 motion vectors and predicting visual dynamics and actions together, at 31.4GB peak memory and 117.7ms latency.

May 16Saturday

Synced · WeChat

Why Robots Need World Models: Top Institutions Release Joint Survey

NTU MARS Lab and collaborators released a 43-page survey on robot world models, covering definitions, architectures, applications, benchmarks, and challenges around action-conditioned consistency, inference efficiency, and physical grounding.

Why it matters: NTU MARS Lab and partner institutions release a 43-page survey on robot world models, covering how they let robots predict physical outcomes before acting, as background on simulation training for embodied AI.

May 14Thursday

Synced · WeChat

China in Focus: PsiBot Uses 100,000 Hours of Human Data for Embodied AI

PsiBot says it uses 100,000 hours of human operation data to train robot policies, with the W0 world model acting only as a training-time transfer module while deployment runs R2 alone.

Why it matters: Lingchu claims 100,000 hours of human operation data trained its robot policy, with the W0 world model used only in training; the data collection and task results can't be confirmed.

May 10Sunday

Synced · WeChat

Turing Award Winner Sutton Uses a 1967 Formula to Improve Streaming Reinforcement Learning

Richard Sutton and coauthors proposed Intentional Updates, which derive the step size from the desired output change; Intentional AC approached SAC on MuJoCo under batch=1 streaming training without replay, while each update used about 1/140 of SAC’s FLOPs.

Why it matters: Sutton's group derives a learning-rate rule from the 1967 Kalman filter, cutting drift in streaming RL that learns one sample at a time with no replay buffer.

May 8Friday

Synced · WeChat

ICLR 2026: NVIDIA and Purdue Use an Agentic Loop for Text-to-3D Scene Generation

NVIDIA Cosmos Lab and Purdue University proposed Scenethesis, a language-and-vision agentic framework for text-to-3D scene generation that uses visual grounding, SDF-based physical constraints, and a judge module; experiments report about 72% first-pass success, 91% after self-checking, and collision rate reduction from 6.1% to 0.8%.

Why it matters: A concrete agent pipeline for 3D scene generation, with SDF physics constraints and a judge module that checks and corrects its own output.

Apr 30Thursday

Synced · WeChat

After Generalist, Jianlan Luo’s Team Releases LWD for Embodied AI Training

Jianlan Luo’s team and Agibot released LWD, tested on 16 Agibot G1 robots in real settings. LWD Online scored 0.95 across 8 tasks and 0.91 on long-horizon tasks. Its offline-to-online RL uses failures as data; failed trajectories were 34.8% of a 652.5-hour pool.

Why it matters: Luo Jianlan's team and AgiBot built the LWD training framework that learns from failure data, and on 16 real Agibot G1 robots it hit 0.95 average success across 8 tasks.

Apr 29Wednesday

X · @dotey

HKUST, NUS, Oxford and others release an 88-page survey on world models

Over 10 universities released an 88-page survey proposing a “capability level × domain law” framework for world models. It reviews 400+ works and reports the best video models pass physical-consistency tests at only 26.2%. The key L3 case is A-Lab: 353 closed-loop experiments in 17 days, yielding 36 compounds.

Why it matters: A 88-page survey from HKUST and a dozen other universities sorts more than 400 world-model papers into one two-axis frame of capability levels and domain laws.

Apr 28Tuesday

QbitAI · WeChat

NTU REI-Bench Tests Vague Human Instructions, With Success Rates Dropping Up to 36.9%

NTU MARS Lab released REI-Bench, a benchmark with 9 ambiguity levels for vague human instructions. Tests used 4 robot planning frameworks and 6 small LLMs; LLaMA3.1-8B+SayCan fell from 57.7% to 46.9% in standard multi-turn context. The key issue is implicit reference resolution, where baseline success dropped 7.4% to 36.9%.

Why it matters: NTU's MARS lab released REI-Bench, grading ambiguous instructions into 9 levels and testing 4 planning frameworks with 6 small models, where one setup's multi-turn success rate fell from 57.7% to 46.9%.

Apr 11Saturday

QbitAI · WeChat

A Chinese embodied model reached global No.1 as a 100,000-hour human dataset for robots was released

Psibot says it released a 100,889-hour human-plus-robot manipulation dataset, and that Psi-R2 ranked first on AllenAI’s MolmoSpace benchmark. The post lists 95,472 hours of human data, 5,417 hours of robot data, 1,000 open-sourced hours, 294 scenes, 4,821 tasks, and 1,382 objects; Psi-W0 adds 30% failure samples, and Psi-R2 latency drops from 2.2s to under 100ms. The key point is the data loop and benchmark framing: the post claims nearly 10x higher success, but does not disclose task setup, full baselines, or statistics.

Why it matters: Psi-R2 tops AllenAI's MolmoSpace benchmark and ships a 100,889-hour manipulation dataset, 95,472 hours of it human, with 1,000 hours open sourced so far.