InternW0: A Foundational Physical World Model for Efficient Real-World Interactions

Jisong Cai; Yao Mu; Ganlin Yang; Zhe Cao; Zhangzheng Tu; Xing Gao; Kailin Li; Xinyu Zhan

cs.ROSep 23, 2026

InternW0: A Foundational Physical World Model for Efficient Real-World Interactions

Authors: Jisong Cai, Yao Mu, Ganlin Yang, Zhe Cao, Zhangzheng Tu, Xing Gao, Kailin Li, Xinyu Zhan,

Organizations: Physical Intelligence Team, Shanghai AI Laboratory

Abstract

Physical intelligence requires more than predicting how the world may evolve: predictions must remain actionable as the world continues to change. We introduce InternW0, the first instantiation of the InternW physical world model series from Shanghai AI Laboratory, built around omnimodal interfaces, asynchronous multi-frequency processing, and local physical modeling under partial observations and external influences. InternW0 jointly learns future visual dynamics and continuous robot control through an asymmetric video--action architecture with flow matching. A high-capacity video expert provides longer-horizon predictive context, while a lightweight action expert operates at a faster timescale. Instead of regenerating the future for every action update, InternW0 reuses layerwise K/V and adapts it to newly observed states through observation-conditioned context routing. Domain-specific interfaces and soft prompts support heterogeneous embodiments, while contact-aware post-training incorporates force and tactile signals for contact-rich manipulation. We train InternW0 on approximately 7,200 hours of heterogeneous robot and egocentric data, including EgoLab, a 275-hour real-laboratory egocentric dataset. Evaluation spans simulation benchmarks and real-world scientific tasks, including a 15-stage metal--organic framework synthesis workflow and 5-stage contact- and force-aware dexterous manipulation for general-purpose quantitative pipetting. These results advance scalable, asynchronous, and science-native physical world models for universal and efficient real-world interactions.

Figures & tables

Figure 1: Framework of InternW world model series.

Figure 2: Architecture of InternW0. Modality-specific video and action experts are coupled through an observation-conditioned chunk K/V editor . The editor adapts cached video context to chunk-level observations, allowing it to be reused across action updates. A frozen Wan VAE and DINOv3 encoder provide visual representations, while precomputed language embeddings condition both experts. Domain-specific interfaces and soft prompts support heterogeneous embodiments; force and tactile channels are introduced optionally during contact-aware post-training.

Dataset	Type	Subsets	#Domains	#Hours	#Episodes
InternData-A1 ( Tian et al., 2026 )	Simulated	Franka (Panda / Robotiq85), Genie1, Lift2, Split ALOHA, AC-One	6	3,494.3	568,194
AgibotWorld ( Bu et al., 2025 )	Real	gripper and tactile subsets	2	2,620.0	158,380
RoboCOIN ( Wu et al., 2025 )	Real	AgiBot G1 (2 subsets), Cobot Magic (5 subsets), Galbot G1, R1 Lite, RMC-AIDA-L, Split ALOHA (2 subsets), Tianqin A2	13	438.1	59,952
Galaxea ( Jiang et al., 2025 )	Real	R1 Lite	1	314.9	15,362
EgoLab	Ego	video-only pathway	1	275.4	3,192
MolmoAct ( Lee et al., 2025 )	Real	YAM bimanual	1	70.3	3,424

Table 1: Pretraining data composition.

Figure 3: EgoLab dataset. Left: example activities with original frames above and hand annotations below. Right: distribution of activity categories.

Method	Spatial	Object	Goal	Long	Average
$\pi_{0}$ ( Black et al., 2024 )	98.0	96.8	94.4	88.4	94.4
InternVLA-M1 ( Chen et al., 2025b )	98.0	99.0	93.8	92.6	95.9
$\pi_{0.5}$ ( Intelligence et al., 2025 )	98.8	98.2	98.0	92.4	96.9
GR00T-N1.7 ( NVIDIA et al., 2025 )	97.7	98.5	97.5	94.4	97.0
OpenVLA-OFT ( Kim et al., 2025 )	97.6	98.4	97.9	94.5	97.1
Fast-WAM ( Yuan et al., 2026 )	98.2	100.0	97.0	95.2	97.6

Table 2: LIBERO Full setting.

Method	Clean (%)	Randomized (%)	Avg. (%)
$\pi_{0}$ ( Black et al., 2024 )	65.92	58.40	62.16
$\pi_{0.5}$ $\pi_{0.5}$ ( Intelligence et al., 2025 )	82.74	76.76	79.75
ABot-M0 ( Yang et al., 2026b )	81.20	80.40	80.80
Motus ( Bi et al., 2026 )	88.66	87.02	87.84
Fast-WAM ( Yuan et al., 2026 )	91.88	91.78	91.83
LingBot-VA ( Li et al., 2026b )	92.90	91.50	92.20

Table 3: RoboTwin 2.0 Full setting.

Method	Clean (%)	Randomized (%)	Avg. (%)
Fast-WAM ( Yuan et al., 2026 )	77.8	1.9	39.9
X-VLA ( Zheng et al., 2026 )	68.0	20.9	44.5
X-WAM ( Guo et al., 2026 )	70.0	25.8	47.9
Spatial Forcing ( Li et al., 2026a )	77.2	26.7	52.0
$\pi_{0.5}$ ( Intelligence et al., 2025 )	70.7	46.0	58.4
4D-WAM ( Yang et al., 2026a )	81.5	41.8	61.6

Table 4: RoboTwin 2.0 Clean2Random setting.

Figure 4: Pretraining data ablation on RoboTwin C2R. Outlines: Clean (left axis); solid fills: Randomized (right axis). Gains are relative to robot-only pretraining (pp).

Figure 5: Five real-world tasks. From left to right: Make Sandwich, Pick Industrial Parts, Sort Tubes, MoF Experiments, and Quantitative Pipetting. The first four are performed on a dual-arm platform, and the last one on a dexterous-hand platform.

Method	Success Rate (%) $\uparrow$			Progress Rate (%) $\uparrow$
Method	Make Sandwich	Pick Industrial Parts	Sort Tubes	MoF Experiments	Quantitative Pipetting
$\pi_{0.5}$	73.3	53.1	86.7	50.2	46.7
Fast-WAM	40.0	50.6	66.7	10.7	18.7
InternW0	73.3	82.7	88.9	68.4	65.3

Table 5: Real-world results on all five tasks. We report success rate (%) for Make Sandwich, Pick Industrial Parts, and Sort Tubes, and progress rate (%) for MoF Experiments and Quantitative Pipetting.

#	Subtask	$\pi_{0.5}$	Fast-WAM	InternW0
1	Pick up funnel from rack	100.0	100.0	100.0
2	Insert funnel into flask	100.0	46.7	100.0
3	Pick up graduated cylinder	100.0	13.3	93.3
4	Pour liquid into flask via funnel	73.3	0.0	80.0
5	Place graduated cylinder back	73.3	0.0	80.0
6	Hold flask steady on platform	73.3	0.0	80.0

Table 6: Subtask success rates and overall task progress rates on MoF Experiments.

#	Subtask	$\pi_{0.5}$	Fast-WAM	InternW0
1	Pickup & reorientation	93.3	46.7	86.7
2	Tip attachment	40.0	13.3	66.7
3	Liquid aspiration	33.3	13.3	66.7
4	Liquid dispensing	33.3	13.3	60.0
5	Tip ejection & return	33.3	6.7	46.7
Progress rate		46.7	18.7	65.3

Table 7: Subtask success rates and overall task progress rates on Quantitative Pipetting task.

Figure 6: Qualitative comparison on the pouring step of MoF Experiments (top) and the tip-attachment step of Quantitative Pipetting (bottom). InternW0 aligns the tool before executing the contact-rich action, whereas the baselines act with visible misalignment.

Figure 7: Critical-path inference efficiency.

World Models for Robotics

World models for robotics predict how scenes and robot states change under actions to support planning, control, and learning.

See all 369 papers

InternW0- $Δ$ : A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data
Sep 25, 2026Xingyu Miao, Zizun Li, Baole Fang +45World Models for Robotics Robot Foundation Models
Physical Object Understanding with a Physically Controllable World Model
May 30, 2026Rahul Venkatesh, Klemen Kotar, Lilian Naing Chen +9Physical Reasoning World Model Learning
Stanford University · OpenAI · Noetik Inc. +1
Riemann-1.0: An Embodied World Action Model for Physical AI
Aug 27, 2026Haofeng Sun, Jiangbo Pei, Fei Kang +13World Models for Robotics Long-Horizon Robotic Manipulation
Riemann Dynamics · Research Intern at Riemann Dynamics

Sep 25, 2026J/K move · Enter open · S save

InternW0- $Δ$ : A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data

Xingyu Miao, Zizun Li, Baole Fang, Kaiwen Song+44

World Models for Robotics Robot Foundation Models Robot Imitation Learning

Open paper AskPDF arXiv

World Action Models (WAMs) jointly model visual dynamics and action generation for generalist robot manipulation. A central challenge is to integrate priors from large-scale pretrained models---including visual dynamics, scene semantics, geometry, and motion---into a unified framework for robot action generation. We introduce InternW0-

Δ

, a unified WAM pretrained on a heterogeneous corpus that outperforms prior methods across simulation benchmarks and real-robot platforms. InternW0-

Δ

combines pretrained visual dynamics, scene-level semantics, 4D geometric and motion priors, and action generation within a Mixture-of-Transformers (MoT) framework. A pretrained video expert and an action expert interact under semantic guidance from a frozen VLM, while a pretrained 4D foundation model injects geometric and motion priors through training-only distillation. We further introduce Causal Imprint, which learns future-relevant scene changes from training-only future supervision and provides predictive representations directly to the action expert without future-video rollout at inference. For large-scale joint training, we construct a heterogeneous corpus of robot demonstrations, UMI data, egocentric human demonstrations, and Ego2Robot data, curated and aligned under a common state-action representation. The resulting corpus contains over 20K hours of processed training data, to our knowledge the largest open-source corpus of its kind. We pretrain InternW0-

Δ

on this corpus and demonstrate strong performance across simulation benchmarks and real-robot platforms. We will open source the training code, model weights, infrastructure, data-processing pipeline, and processed data where licenses permit. Project page: https://internrobotics.github.io/InternW0-Delta/

Abstract

World Models for Robotics

Explore similar work

InternW0-ΔΔΔ: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data

Physical Object Understanding with a Physically Controllable World Model

Riemann-1.0: An Embodied World Action Model for Physical AI

InternW0- $Δ$ : A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data