MoLA infers a mixture of latent actions from generated future videos via modality-aware inverse dynamics models to improve robot manipulation policies.
hub Canonical reference
Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planning
Canonical reference. 78% of citing Pith papers cite this work as background.
43
Pith papers citing it
Background
78% of classified citations
abstract
Recent video generation models demonstrate remarkable ability to capture complex physical interactions and scene evolution over time. To leverage their spatiotemporal priors, robotics works have adapted video models for policy learning but introduce complexity by requiring multiple stages of post-training and new architectural components for action generation. In this work, we introduce Cosmos Policy, a simple approach for adapting a large pretrained video model (Cosmos-Predict2) into an effective robot policy through a single stage of post-training on the robot demonstration data collected on the target platform, with no architectural modifications. Cosmos Policy learns to directly generate robot actions encoded as latent frames within the video model's latent diffusion process, harnessing the model's pretrained priors and core learning algorithm to capture complex action distributions. Additionally, Cosmos Policy generates future state images and values (expected cumulative rewards), which are similarly encoded as latent frames, enabling test-time planning of action trajectories with higher likelihood of success. In our evaluations, Cosmos Policy achieves state-of-the-art performance on the LIBERO and RoboCasa simulation benchmarks (98.5% and 67.1% average success rates, respectively) and the highest average score in challenging real-world bimanual manipulation tasks, outperforming strong diffusion policies trained from scratch, video model-based policies, and state-of-the-art vision-language-action models fine-tuned on the same robot demonstrations. Furthermore, given policy rollout data, Cosmos Policy can learn from experience to refine its world model and value function and leverage model-based planning to achieve even higher success rates in challenging tasks. We release code, models, and training data at https://research.nvidia.com/labs/dir/cosmos-policy/
hub tools
citation-role summary
background 19
baseline 2
method 2
citation-polarity summary
years
2026 43representative citing papers
NoiseGate learns per-latent timestep schedules as an information-gating policy in diffusion-based world action models, yielding consistent gains on RoboTwin manipulation tasks.
OA-WAM uses persistent address vectors and dynamic content vectors in object slots to enable addressable world-action prediction, improving robustness on manipulation benchmarks under scene changes.
EA-WM generates more accurate robot world rollouts by projecting actions as structured visual fields in camera space and using event-aware bidirectional fusion to better capture interaction dynamics.
Being-H0.7 adds future-aware latent reasoning to direct VLA policies via dual-branch alignment on latent queries, matching world-model benefits at VLA efficiency.
CF-VLA uses a coarse initialization over endpoint velocity followed by single-step refinement to achieve strong performance with low inference steps on CALVIN, LIBERO, and real-robot tasks.
Action Images turn robot arm motions into interpretable multiview pixel videos, letting video backbones serve as zero-shot policies for end-to-end robot learning.
JailWAM is the first dedicated jailbreak framework for World Action Models, achieving 84.2% attack success rate on LingBot-VA in RoboTwin simulation and enabling safety evaluation of robotic AI.
PlayWorld learns high-fidelity robot world models from unsupervised self-play, producing physically consistent video predictions that outperform models trained on human data and enabling 65% better real-world policy performance via model-based RL.
DiM-WAM is a memory-augmented world-action model that integrates multi-scale historical events and global task progress to improve long-horizon robot manipulation performance.
SafeVLA-Bench adds STL-based safety checks to VLA benchmarks and finds 13-56% of successful rollouts on LIBERO and RoboCasa-365 violate at least one safety clause.
VLAConf is a one-class discriminative method that estimates step-wise task-success confidence for VLA models via anomaly scoring on frozen representations plus step-conditioned modeling, shown to be more efficient than ensemble or probability baselines on LIBERO and real robots.
DeMiAn re-annotates robot and egocentric videos with VLM-generated dense labels across motion, scene, pose, and reasoning aspects, then uses a learned instructor to boost policy success by 5 points on RoboCasa over task-only baselines.
OmniHumanoid factorizes transferable motion learning from embodiment-specific adaptation to enable scalable cross-embodiment video generation without paired data for new humanoids.
HarmoWAM unifies predictive and reactive control in world action models via an adaptive gating mechanism to deliver improved zero-shot generalization and precision in robotic manipulation.
DALI-R boosts 3D imitation learning success rates by 6.8% on average from suboptimal trajectories via latent imagination and reranking, with under 0.7x inference cost.
Anchor-Centric Adaptation escapes the diversity trap by prioritizing repeated demonstrations at core anchors over broad coverage, yielding higher success rates under fixed data budgets in robotic manipulation.
MolmoAct2 is an open VLA model that outperforms baselines like Pi-05 on 7 benchmarks and whose backbone surpasses GPT-5 on 13 embodied-reasoning tasks through new datasets, specialized training, and architecture changes for lower latency.
MotuBrain jointly models video and action via a three-stream Mixture-of-Transformers UniDiffuser to reach 95.8-96.1% success on RoboTwin 2.0 benchmarks, top EWMScore, and fast 11 Hz inference while adapting to new robots with 50-100 trajectories.
X-WAM unifies robotic action execution and 4D world synthesis by adapting video diffusion priors with a lightweight depth branch and asynchronous noise sampling, achieving 79-91% success on robot benchmarks.
The paper introduces a unified framework for world models that fully incorporates all cognitive functions from Cognitive Architecture Theory, highlights under-researched areas in motivation and meta-cognition, and proposes Epistemic World Models as a new category for scientific discovery agents.
XRZero-G0 enables 2000-hour robot-free datasets that, when mixed 10:1 with real-robot data, match full real-robot performance at 1/20th the cost and support zero-shot transfer.
Vision-geometry backbones using pretrained 3D world models outperform vision-language and video models for robotic manipulation by enabling direct mapping from visual input to geometric actions.
VAG is a synchronized dual-stream flow-matching framework that generates aligned video-action pairs for synthetic embodied data synthesis and policy pretraining.
citing papers explorer
-
From Imagined Futures to Executable Actions: Mixture of Latent Actions for Robot Manipulation
MoLA infers a mixture of latent actions from generated future videos via modality-aware inverse dynamics models to improve robot manipulation policies.
-
NoiseGate: Learning Per-Latent Timestep Schedules as Information Gating in World Action Models
NoiseGate learns per-latent timestep schedules as an information-gating policy in diffusion-based world action models, yielding consistent gains on RoboTwin manipulation tasks.
-
OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulation
OA-WAM uses persistent address vectors and dynamic content vectors in object slots to enable addressable world-action prediction, improving robustness on manipulation benchmarks under scene changes.
-
EA-WM: Event-Aware Generative World Model with Structured Kinematic-to-Visual Action Fields
EA-WM generates more accurate robot world rollouts by projecting actions as structured visual fields in camera space and using event-aware bidirectional fusion to better capture interaction dynamics.
-
Being-H0.7: A Latent World-Action Model from Egocentric Videos
Being-H0.7 adds future-aware latent reasoning to direct VLA policies via dual-branch alignment on latent queries, matching world-model benefits at VLA efficiency.
-
CF-VLA: Efficient Coarse-to-Fine Action Generation for Vision-Language-Action Policies
CF-VLA uses a coarse initialization over endpoint velocity followed by single-step refinement to achieve strong performance with low inference steps on CALVIN, LIBERO, and real-robot tasks.
-
Action Images: End-to-End Policy Learning via Multiview Video Generation
Action Images turn robot arm motions into interpretable multiview pixel videos, letting video backbones serve as zero-shot policies for end-to-end robot learning.
-
JailWAM: Jailbreaking World Action Models in Robot Control
JailWAM is the first dedicated jailbreak framework for World Action Models, achieving 84.2% attack success rate on LingBot-VA in RoboTwin simulation and enabling safety evaluation of robotic AI.
-
PlayWorld: Learning Robot World Models from Autonomous Play
PlayWorld learns high-fidelity robot world models from unsupervised self-play, producing physically consistent video predictions that outperform models trained on human data and enabling 65% better real-world policy performance via model-based RL.
-
DIM-WAM: World-Action Modeling with Diverse Historical Event Memory
DiM-WAM is a memory-augmented world-action model that integrates multi-scale historical events and global task progress to improve long-horizon robot manipulation performance.
-
SafeVLA-Bench: A Benchmark for the Success-Safety Gap in Vision-Language-Action Models
SafeVLA-Bench adds STL-based safety checks to VLA benchmarks and finds 13-56% of successful rollouts on LIBERO and RoboCasa-365 violate at least one safety clause.
-
VLAConf: Calibrated Task-Success Confidence for Vision-Language-Action Models
VLAConf is a one-class discriminative method that estimates step-wise task-success confidence for VLA models via anomaly scoring on frozen representations plus step-conditioned modeling, shown to be more efficient than ensemble or probability baselines on LIBERO and real robots.
-
How to Instruct Your Robot: Dense Language Annotations Power Robot Policy Learning
DeMiAn re-annotates robot and egocentric videos with VLM-generated dense labels across motion, scene, pose, and reasoning aspects, then uses a learned instructor to boost policy success by 5 points on RoboCasa over task-only baselines.
-
OmniHumanoid: Streaming Cross-Embodiment Video Generation with Paired-Free Adaptation
OmniHumanoid factorizes transferable motion learning from embodiment-specific adaptation to enable scalable cross-embodiment video generation without paired data for new humanoids.
-
HarmoWAM: Harmonizing Generalizable and Precise Manipulation via Adaptive World Action Models
HarmoWAM unifies predictive and reactive control in world action models via an adaptive gating mechanism to deliver improved zero-shot generalization and precision in robotic manipulation.
-
Data-Asymmetric Latent Imagination and Reranking for 3D Robotic Imitation Learning
DALI-R boosts 3D imitation learning success rates by 6.8% on average from suboptimal trajectories via latent imagination and reranking, with under 0.7x inference cost.
-
Escaping the Diversity Trap in Robotic Manipulation via Anchor-Centric Adaptation
Anchor-Centric Adaptation escapes the diversity trap by prioritizing repeated demonstrations at core anchors over broad coverage, yielding higher success rates under fixed data budgets in robotic manipulation.
-
MolmoAct2: Action Reasoning Models for Real-world Deployment
MolmoAct2 is an open VLA model that outperforms baselines like Pi-05 on 7 benchmarks and whose backbone surpasses GPT-5 on 13 embodied-reasoning tasks through new datasets, specialized training, and architecture changes for lower latency.
-
MotuBrain: An Advanced World Action Model for Robot Control
MotuBrain jointly models video and action via a three-stream Mixture-of-Transformers UniDiffuser to reach 95.8-96.1% success on RoboTwin 2.0 benchmarks, top EWMScore, and fast 11 Hz inference while adapting to new robots with 50-100 trajectories.
-
Unified 4D World Action Modeling from Video Priors with Asynchronous Denoising
X-WAM unifies robotic action execution and 4D world synthesis by adapting video diffusion priors with a lightweight depth branch and asynchronous noise sampling, achieving 79-91% success on robot benchmarks.
-
Human Cognition in Machines: A Unified Perspective of World Models
The paper introduces a unified framework for world models that fully incorporates all cognitive functions from Cognitive Architecture Theory, highlights under-researched areas in motivation and meta-cognition, and proposes Epistemic World Models as a new category for scientific discovery agents.
-
XRZero-G0: Pushing the Frontier of Dexterous Robotic Manipulation with Interfaces, Quality and Ratios
XRZero-G0 enables 2000-hour robot-free datasets that, when mixed 10:1 with real-robot data, match full real-robot performance at 1/20th the cost and support zero-shot transfer.
-
Robotic Manipulation is Vision-to-Geometry Mapping ($f(v) \rightarrow G$): Vision-Geometry Backbones over Language and Video Models
Vision-geometry backbones using pretrained 3D world models outperform vision-language and video models for robotic manipulation by enabling direct mapping from visual input to geometric actions.
-
VAG: Dual-Stream Video-Action Generation for Embodied Data Synthesis
VAG is a synchronized dual-stream flow-matching framework that generates aligned video-action pairs for synthetic embodied data synthesis and policy pretraining.
-
Grasp as You Dream: Imitating Functional Grasping from Generated Human Demonstrations
GraspDreamer synthesizes human functional grasping demonstrations with visual generative models to enable zero-shot robot grasping with improved data efficiency and generalization.
-
Veo-Act: How Far Can Frontier Video Models Advance Generalizable Robot Manipulation?
Veo-3 video predictions enable approximate task-level robot trajectories in zero-shot settings but require hierarchical integration with low-level VLA policies for reliable manipulation performance.
-
Fast-WAM: Do World Action Models Need Test-time Future Imagination?
Fast-WAM shows that explicit future imagination at test time is not required for strong WAM performance; video modeling during training provides the main benefit.
-
World Action Models are Zero-shot Policies
DreamZero uses a 14B video diffusion model as a World Action Model to achieve over 2x better zero-shot generalization on real robots than state-of-the-art VLAs, real-time 7Hz closed-loop control, and cross-embodiment transfer with 10-30 minutes of data.
-
PALM: Progress-Aware Policy Learning via Affordance Reasoning for Long-Horizon Robotic Manipulation
PALM improves long-horizon robotic manipulation success by distilling affordance representations for object interaction and predicting within-subtask progress in a VLA model.
-
$\tau_0$-WM: A Unified Video-Action World Model for Robotic Manipulation
A shared video diffusion backbone jointly predicts future latents and continuous actions while also rolling out candidate actions to predict dense task-progress scores, trained on 27,300 hours of mixed robot and human data.
-
World Models for Robotic Manipulation: A Survey
Survey organizing world models for robotic manipulation into representation families, a functional taxonomy, and infrastructure roles across pretraining, post-training, and inference, while reviewing 34 datasets and evaluation protocols.
-
SANTS: A State-Adaptive Scheduler for World Action Models
SANTS adaptively chooses denoising depth in video-based robot action diffusion policies using a state-dependent stopping hazard and noise ratio, trained via downstream action reward to reduce latency.
-
PhyWorld: Physics-Faithful World Model for Video Generation
PhyWorld improves temporal consistency and physical plausibility in video world models via flow matching fine-tuning followed by DPO on physics preference pairs, with reported gains on VBench and a custom physical-faithfulness benchmark.
-
Nautilus: From One Prompt to Plug-and-Play Robot Learning
NAUTILUS is a prompt-driven harness that automates plug-and-play adapters, typed contracts, and validation for policies, benchmarks, and robots in learning research.
-
Gated Memory Policy
GMP selectively activates and represents memory via a gate and lightweight cross-attention, yielding 30.1% higher success on non-Markovian robotic tasks while staying competitive on Markovian ones.
-
Causal World Modeling for Robot Control
LingBot-VA combines video world modeling with policy learning via Mixture-of-Transformers, closed-loop rollouts, and asynchronous inference to improve robot manipulation in simulation and real settings.
-
World Action Models: The Next Frontier in Embodied AI
The paper introduces World Action Models as a new paradigm unifying predictive world modeling with action generation in embodied foundation models and provides a taxonomy of existing approaches.
-
RLDX-1 Technical Report
RLDX-1 outperforms frontier VLAs such as π0.5 and GR00T N1.6 on dexterous manipulation benchmarks, reaching 86.8% success on ALLEX humanoid tasks versus around 40% for the baselines.
-
Matrix-Game 3.0: Real-Time and Streaming Interactive World Model with Long-Horizon Memory
Matrix-Game 3.0 delivers 720p real-time video generation at 40 FPS with minute-scale memory consistency by combining residual self-correction training, camera-aware memory injection, and DMD-based autoregressive distillation on a 5B model.
-
World Models: A Comprehensive Survey of Architectures, Methodologies, Reasoning Paradigms, and Applications
The paper delivers a multi-axis taxonomy for world models that maps architectures, training families, reasoning strategies, and domains from early cognitive foundations through systems such as Dreamer, MuZero, and Sora while noting evaluation gaps.
-
World Model for Robot Learning: A Comprehensive Survey
A comprehensive survey that organizes the literature on world models in robot learning, their roles in policy learning, planning, simulation, and video-based generation, with connections to navigation, driving, datasets, and benchmarks.
- Sensor2Sensor: Cross-Embodiment Sensor Conversion for Autonomous Driving
- VideoGPA: Distilling Geometry Priors for 3D-Consistent Video Generation