World Models for VLAs
VLA: Vision-Language-Action Model
Latest papers
65 papers28 papers in the last four weeks, up 155% on the four weeks before. 0.2% of all new papers.
We introduce -sight, a unified robot foundation model that couples vision-language-action (VLA) and generative latent world modeling. Video-based world action models (WAMs) show the value of joint future-and-action generation for control, yet inherit the heavy cost of video generators. Recent latent-space policies avoid pixel generation, yet typically align intermediate features or rely on frozen target encoders, decoupling future prediction from the VLM's native feature space. -sight instead predicts the future directly in the visual token space in which its VLM perceives the present. Coupling a foresight expert and a continuous action expert with the VLM in a mixture-of-transformers, -sight jointly denoises future spatial tokens and actions at inference without pixel reconstruction or external encoders. Trained on human video and robot trajectories, -sight achieves 81.9% success on SIMPLE (+9.6 points over the strongest baseline) and 73.3% across six Unitree G1 tasks (+20.0 points) at 110 ms per chunk, while remaining competitive on tabletop benchmarks.
TrajWAM: Trajectory-Conditioned Future Learning for Vision-Language Driving
World Action Models (WAMs) connect future scene prediction with action planning for autonomous driving. Each driving record pairs an executed ego trajectory with its observed future, providing a factual outcome for learning this connection. We present TrajWAM, a trajectory-conditioned WAM that improves vision-language driving through auxiliary bird's-eye-view (BEV) future learning. An ego-trajectory query conditions DINOv3-aligned BEV predictions from shared planner representations. Training combines factual reconstruction under the executed trajectory with one-sided feature separation under an unlabeled perturbed trajectory. Together, these objectives train the shared planner to generate driving waypoints. Native inference directly generates waypoints and bypasses the auxiliary future-prediction branch. On all 220 Bench2Drive routes, TrajWAM achieves 91.53 Driving Score and 75.45% Success Rate, exceeding a schedule-matched planning-only fine-tuning baseline by 5.31 Driving Score points and eight successful routes.
JEPA Dots: Dual-Latent World Model for Language-Guided Robot Control
Predictive world models for language-guided robot control face a fundamental conditioning challenge: instructions should influence action selection without altering the predicted consequences of fixed actions under identical observations. To address this challenge, we introduce JEPA Dots (Dynamics-Oriented Task Separation), a compact dual-latent framework that jointly learns visual dynamics and language-guided control through separate dynamics and task representations. Language-conditioned task features guide a flow-matching policy, while masked attention restricts world prediction to instruction-independent dynamics history and supplied actions. In closed-loop evaluation on 40 trained LIBERO tasks, JEPA Dots achieves 84.9% overall success with approximately 39.1M parameters, compared with the published 78.7% of NanoVLA-S with 161.0M parameters under a different evaluation protocol. In the reference experiment, JEPA Dots also improves overall success from 76.0% for our task-ID-conditioned adaptation baseline to 84.9%, with LIBERO-Goal success increasing from 67.1% to 88.5%. Instruction-switching experiments further confirm that task representations and policy outputs respond to language changes, while dynamics representations and world predictions remain invariant under fixed observations and supplied actions.
LaST1.0: A Unified Latent World-Action Model with Control-Centric SpatioTemporal Reasoning
World-Action Models couple action learning with future prediction, enabling robotic policies to capture environmental dynamics for robust control. However, prevailing methods take a reconstruction-oriented approach to model future videos, allocating substantial capacity to appearance details that are incidental to control. In this work, we reformulate world modeling for manipulation as control-centric future prediction in latent space, and present LaST1.0, a unified latent World-Action Model. Rather than modeling an observation-complete video future, LaST1.0 predicts compact, geometry-aware latent targets derived from future wrist views, focusing the supervision on the evolving robot-object interactions required for fine-grained manipulation. To better support this control-centric paradigm, LaST1.0 revisits the visual input interface and attention mechanism of its unified architecture. It adopts a vision-encoder-free design with a lightweight patch embedder, exposing fine-grained visual cues directly to the backbone, and introduces LaST-Attention for structured spatiotemporal interactions across modalities. Evaluations across two robot embodiments, five challenging manipulation tasks, and three simulation benchmarks---RLBench, LIBERO, and LIBERO-Plus---demonstrate that LaST1.0 achieves state-of-the-art success rates and generalizes to novel scenarios. These results demonstrate the effectiveness of control-centric latent world modeling for robust manipulation.
CoT-WAM: Collaborative Training of Vision-Language Models and World-Action Models via Plan Latents
General robot manipulation requires both semantic competence and physical foresight. World action models (WAMs) acquire an understanding of and predictive capability over the physical world by jointly modeling visual foresight and action distributions. However, existing WAMs still lack semantic competence: they have limited ability to perform open-world scene understanding, semantic reasoning, and long-horizon planning. Although vision-language models (VLMs) provide these capabilities, effectively integrating their semantic priors into WAM prediction and control remains challenging. To address this challenge, we propose CoT-WAM, a jointly trained WAM-VLM framework. CoT-WAM uses a learnable embedding to exchange information between the VLM and the WAM, which is supervised by textual and goal-image signals during training to encode rich intent information. A tailored two-stage training strategy enables joint optimization and more effective collaboration between the VLM and WAM. We evaluate CoT-WAM on six benchmarks covering control performance, zero-shot generalization, and long-horizon planning. CoT-WAM achieves SOTA performance on LIBERO, CALVIN, and RoboCasa-Kitchen, and demonstrates stronger zero-shot generalization on LIBERO-Plus and LIBERO-Pro without fine-tuning. On RoboCasa365, it further improves long-horizon planning over baselines with and without external planners. Real-world experiments further demonstrate the effectiveness of our method in practical deployments.
Making Future-Predictive Representations Actionable
Recent world action models (WAMs) drop future prediction at inference, assuming that representations learned through auxiliary future supervision still benefit action generation. However, our layerwise analysis of Fast-WAM reveals a depth-dependent mismatch: while probes decode future state transitions more accurately from deeper visual layers, the gradients and visual focuses of the future and action objectives increasingly diverge with depth, leaving much of this predictive information unused by the action pathway. To address this issue, we introduce RelayWAM, which adds an explicit transition space between the video and action backbones. Conditioned on the task instruction, robot state, and current action estimate, recurrent transition layers extract step-relevant predictive cues across depth and relay them directly into the action attention, without generating future frames at inference. Experiments show that RelayWAM substantially amplifies the benefit of future supervision, translating into stronger generalization under distribution shifts in both simulation and real-world robot tasks. On LIBERO-Plus, RelayWAM achieves 72.03% mean success, compared to 56.42% without future supervision and 51.50% for Fast-WAM.
Learning Agent-Centric Vision-Language-Action World Models for Autonomous Driving
Recent progress in integrating vision-language-action (VLA) models with world modeling has significantly advanced end-to-end autonomous driving by allowing systems to reason about imagined futures instead of merely reacting to current observations. However, most world VLA models generate future frames at the pixel level with plenty of dense visual tokens. Although such generated futures can be visually plausible, they preserve many driving-irrelevant appearance details, i.e., only a small portion of the tokens directly encode critical planning-relevant factors, such as agent geometry, motion dynamics, and interactions. This introduces a significant performance bottleneck. An important question thus arises: what world state should a world VLA model reason over to support planning beyond pixel-level prediction? We argue that planning-oriented world modeling should focus on a compact and structured agent-centric latent state that captures the interactions between objects in the driving scene over time. As such, we introduce AgentWorld, an agent-centric VLA world model that uses discrete agent tokens jointly for reasoning and trajectory planning. Each token is dynamically grounded to a surrounding agent associated with its object's state, such as semantic identity, 3D geometry, motion, and interaction context. AgentWorld learns these tokens through three key processes: explicit geometric grounding, reasoning-oriented supervised fine-tuning, and agent-oriented reinforcement learning. Extensive open-loop and closed-loop experiments under diverse driving scenarios show that AgentWorld enhances planning capabilities, fully exploits visual evidence and agent dynamics, and yields interpretable agent-centric world states.
ReLAP: Latent Action Pretraining through Recurrent Video Prediction in JEPA Space
Videos provide a rich source of physical interaction for learning robot behavior without action annotations, yet effective latent actions must capture not only local transitions but also how those transitions evolve over time. We introduce ReLAP, a latent-action pretraining framework that couples local transition learning with recurrent multi-step prediction in a frozen JEPA representation space. A shared inverse dynamics model infers compact continuous latent actions from adjacent observations, while a latent world model recurrently predicts the future JEPA features. Because each predicted feature becomes the context for the next transition, errors at later steps backpropagate through the rollout to earlier latent actions, encouraging temporally coherent dynamics beyond pairwise reconstruction. We transfer this representation to a vision-language model using two complementary objectives: latent distillation from the frozen inverse teacher and recurrent future-feature prediction through the frozen latent world model. The resulting policy is then adapted to robot demonstrations with action flow matching. At deployment, the explicit latent-action and world-model branches are removed, and the fine-tuned VLM predicts physical actions directly, reducing inference complexity. With only 11M pretraining samples and 1.06B trainable parameters, ReLAP achieves 97.9% on standard LIBERO and 70.0% over zero-shot LIBERO-Plus perturbations. On the LARY latent-action benchmark, ReLAP ranks second in both mean regression error and mean classification accuracy among the compared methods. These results demonstrate that recurrent action-free latent action pretraining can learn compact, temporally structured, and transferable representations for downstream robot control.
ThinkWAM: Rethinking Latent Futures for Action Refinement
World-action models (WAMs) couple action generation with predictions of future states. An initial action chunk may be flawed. A future prediction that is not conditioned on the chunk is identical for every proposal and thus carries no proposal-specific signal, limiting its usefulness for refinement before execution. We introduce ThinkWAM, a latent WAM that establishes bidirectional feedback before execution through a proposal-rethink-refine loop: the policy first proposes a complete action chunk, the proposal corrects multi-horizon future latents through a learned residual, and the corrected latents are injected into a second pass of the same action head, which generates the executable chunk. An affordance-guided observation-action memory and a near-fine, far-coarse multi-horizon prediction scheme, both motivated by predictive coding, support this loop, and all prediction stays in latent space. ThinkWAM achieves 98.70% success on LIBERO, 84.76% on LIBERO-Plus without OOD fine-tuning, 85.75%/65.85% on RoboTwin 2.0 Clean/Random, and 65.17% on RoboCasa24. ThinkWAM achieves 64.4% mean success across three real-world OOD conditions. The mean two-pass inference latency per action chunk is 153.48 ms on LIBERO, 164.61 ms on RoboTwin, and 101.2 ms on the Galaxea A1Z. Ablations show complementary gains from rethinking and affordance guidance. Diagnostics indicate that proposal-conditioned corrections change both predicted futures and action trajectories, showing that the corrected future representations improve the final action.
DiffWAM: Learning World Action Model from Consecutive Visual Difference
World Action Models (WAMs) have emerged as an important paradigm in robot learning, jointly predicting future scene states and robot actions from current observations and instructions. However, future states entangle task-related changes with large amounts of task-irrelevant content, such as static backgrounds, which weakens the model's ability to capture changes relevant to task execution. To address this, we propose Difference World Action Model (DiffWAM), a Mixture-of-Experts (MoE) framework that enhances world representations by jointly learning future-state prediction and consecutive visual difference prediction. We define consecutive visual differences as differences between the latent representations of consecutive observations, providing an explicit learning signal for temporal variation. The framework comprises a World Expert for future frame prediction, an Action Expert for action generation, and a Diff Expert for predicting these latent differences along the trajectory. A structured unidirectional attention mask conditions the Diff, World, and Action Experts solely on the current-frame representation while blocking attention between experts. This design allows both the World and Diff Experts to be removed during inference, introducing no additional inference overhead. DiffWAM achieves state-of-the-art performance on the LIBERO and RoboTwin 2.0 benchmarks and substantially improves over FastWAM on real-world manipulation tasks. Specifically, it attains average success rates of 98.8%, 92.65%, and 88.3%, outperforming the state-of-the-art baseline (i.e., FastWAM) by 1.75, 1.55, and 15.0 percentage points, respectively.
ResWAM: Learning What Changes, Not What Stays, for World-Action Models
World-action models (WAMs) augment vision-language-action policies (VLAs) with future prediction so that the policy can anticipate how its actions will affect the physical world. Existing WAMs supervise complete future observations. However, observations typically change little over an action chunk, so the future loss is dominated by copying static background content rather than learning task-relevant robot and object motion, limiting policy learning. Moreover, existing WAMs rely on large video generation models or separate world models, incurring heavy training and inference compute that hinders closed-loop deployment. To improve policy performance, we propose visual residual learning, supervising only the feature change from the current frame to the future frame in a frozen DINOv3 space. Compared with complete future features, this residual learning target yields a higher feature rank with richer and more diverse representations of scene changes, and spatially concentrates on the robot arm and manipulated objects rather than the background. To further improve computational efficiency, we present ResWAM, a residual world-action model with 2.3B parameters in total, including a lightweight 0.5B dual-branch DiT that jointly trains an action branch and a residual branch via conditional flow matching. Without robot pretraining (i.e., pretraining on external robot action trajectories beyond the benchmark demonstrations), ResWAM achieves top average success rates on LIBERO-Plus (81.1%), RoboTwin 2.0 (93.8%), and RoboDojo (9.40% success, 13.41 score) at the lowest training (2.7-4.0 TFLOPs) and inference compute (1.0-1.4 TFLOPs), and leads real-robot Open, Put, and Stack (80.0% average, +10.0% over ) at 92.7 ms latency.
Vision-Language-Action Models: A Life-Cycle Survey of Design, Training and Deployment
Vision-language-action (VLA) models adapt web-pretrained vision-language models to output robot actions, so that web knowledge carries over to physical control. In under 4 years the field has moved from RT-2, the first VLA, to open models that fine-tune on a single consumer GPU and to proprietary generalists that clean unseen homes. This survey reviews 88 works from December 2022 to October 2026, mainly on manipulation and navigation, plus humanoid control, world-action models and evaluation. We follow a VLA through its life cycle, from model design to training and deployment, along seven axes. Model design covers (a) inputs and memory, what the model observes; (b) architecture, how its pipeline is built; (c) action output, how it produces motor commands; and (d) intermediate predictions, what else it predicts, from reasoning steps to future observations. Training covers (e) pretraining data, the mix of robot, web, human and simulated data, and (f) post-training, adaptation to the target robot. Deployment covers (g) real-time execution, keeping pace with the robot. We analyse 22 models on these axes and report their training data, model size, inference speed and hardware. Three trends emerge: (i) training data has broadened, from about 10,000 hours of teleoperated robot data () to 23,000 community-collected episodes (SmolVLA) and 20,000 hours of egocentric human video (GR00T N1.7); (ii) model sizes have split: most recent models have 3-7 billion parameters, while a new group under one billion runs on consumer GPUs; and (iii) generalist VLAs are emerging that handle both manipulation and navigation, such as UniVLA and Qwen-VLA. We also compare VLAs with world-action models, which predict future observations as well as actions. We review evidence that some VLAs with near-perfect benchmark scores fail when objects or the camera move slightly, and open up further research questions.
Bind-WAM: Binding-Uncertainty-Guided Selective Future Imagination in World Action Models
World Action Models (WAMs) can improve robot learning by jointly predicting future visual states and actions, but generating a visual future at every replanning step is expensive and sometimes unnecessary. This raises a critical question: when should a robot policy imagine the future? In this paper, we propose Bind-WAM, a binding-uncertainty-guided WAM for selective future imagination. Bind-WAM introduces language-object binding uncertainty, selectively invoking joint future-action generation when the current observation does not support stable task-relevant action generation. An offline labeling pipeline uses complete demonstrations and action context to identify task-relevant objects and detect reference mismatch, invisibility, and identity drift. The resulting pseudo-labels supervise a lightweight router that predicts binding uncertainty from the current observation and language instruction. At inference, a shared WAM policy invokes selective future imagination according to the predicted uncertainty and its change between consecutive replanning steps. Simulation and real-world experiments demonstrate that Bind-WAM can establish binding uncertainty as a strong, interpretable signal for selective future imagination.
From Video Frames to Robot Meta-Frames: A Unified Latent Interface for World Action Models
Pretrained video diffusion models (VDMs) offer a powerful foundation for World Action Models (WAMs), which jointly predict robot actions and future observations. Building such models requires injecting low-dimensional robot signals (e.g. proprioception, actions, and values) into a latent space shaped by natural videos, and a key design choice is the interface through which these signals enter: existing schemes treat it as a formatting detail, spending a separate latent frame on each signal or bolting on dedicated encoders and decoders. We argue that signal injection is instead a distribution-alignment problem: how well a representation matches the prior of the pretrained VDMs during denoising. Using the relative reconstruction error of a frozen VDM denoiser as a probe, we find that conventional spatial encodings sit far off this prior, whereas frequency-domain packing yields latents that the denoiser corrects far less. Motivated by this diagnostic result, we propose MetaWAM, which folds all robot signals into a single unified meta-frame: each signal is assigned to a disjoint frequency-domain channel band and lifted to the spatial latent by an inverse DCT, yielding a decodable interface aligned with the pretrained VDM prior without any added module or architectural change. A fixed band layout together with slot-aware routing lets the same meta-frame support policy, world-model, and value prediction over a shared set of robot variables. On RoboCasa, MetaWAM reaches 68.9% success while collapsing the per-signal frames into one and cutting inference latency by 39%, and attains 85.0% success on three real-world Franka tasks. Code and models are available at https://github.com/showlab/MetaWAM
SWAM: Scaling World Action Models with Cross-Modal Alignment
World action models jointly predict video and actions conditioned on language. Existing studies focus primarily on architectures and training mechanisms, typically treating language-video alignment acquired through video pretraining as an established foundation. However, introducing action learning may impair this alignment. Principled guidance on data and supervision design for maintaining language-video alignment during action learning remains limited. To investigate this, we curate a pool of approximately 47,000 hours of general videos, egocentric videos, UMI demonstrations, and robot demonstrations. Building on this, we systematically study how language-video supervision, language-video-action supervision, and the joint use of these two forms affect WAM performance. In particular, we examine how language annotation quality affects pretrained language-video capabilities, and how action supervision from UMI and robot demonstrations extends these capabilities to language-video-action modeling. To better scale the model, we further conduct detailed studies on joint training recipe and model architecture. Under a fixed training budget, mixing egocentric videos and UMI demonstrations improves control generalization. Model architecture affects how effectively these experiences are used, while scaling pretraining under the selected configuration brings further gains. The resulting recipe, post-trained on RoboTwin Clean demonstrations only, achieves success rates of 88.10% and 61.56% under Easy (Clean-to-Clean) and Hard (Clean-to-Random) conditions, respectively. Our model achieves 40.00% SR across three real-world tasks, compared with 23.33% for the strongest baseline, indicating better generalization under unseen conditions. We release the curated data, training code, and model weights.
ForeAct: Geometry-Guided Action Learning from World Representations
Robotic manipulation requires representations that capture both scene geometry and its evolution during interaction. Existing approaches have largely treated these two aspects separately. World action models capture such dynamics by coupling action prediction with future video generation, but their reliance on iterative diffusion or rectified-flow inference leads to high computational cost, while their representations lack explicit spatial structure. Explicitly modeling scene geometry typically requires auxiliary modules that increase both parameter count and computational cost. Conversely, geometry-centric policies primarily recover observed spatial structure and often lack explicit supervision for anticipating future geometric changes during interaction. To address these limitations, we introduce ForeAct, which repurposes a pretrained video world model as a single-pass representation backbone for low-latency control. ForeAct directly processes the clean latent of the current observation in one forward pass, eliminating future-frame synthesis and iterative denoising while retaining the backbone's pretrained spatiotemporal priors. Rather than introducing a separate geometry backbone, ForeAct uses learnable tokens supervised by next-frame relative depth to shape the shared video representation through bidirectional self-attention. This compact interface turns the spatial knowledge already encoded in the backbone into a future-aware representation for action generation, with little parameter overhead. The lightweight geometry readout is used only during training and removed at deployment, enabling low-latency, geometry-aware control without depth input or an external geometry backbone. Without additional large-scale pretraining, ForeAct achieves 96.6% average success on LIBERO, 63.5% on SimplerEnv-Bridge, and a 74.4% success rate across 90 real-world rollouts.
DexForesight: Grounding Dexterous Policies with Implicit Spatial Foresight
Dexterous manipulation requires rich spatial foresight over both the surrounding scene and the robot hand. Existing world-action models often acquire such predictive knowledge through explicit future prediction. We introduce DexForesight, a framework that learns spatial foresight implicitly by injecting dynamics-aware world representations into the latent space of a dexterous policy. During training, DexForesight learns an action-conditioned spatial world model from offline scene depth, kinematic hand geometry, and demonstration trajectories. The world model predicts 3D hand-point motion and visual feature changes, capturing both articulated hand dynamics and scene changes under demonstrated actions, including the motion of fingers occluded from the cameras. We then freeze the world model and align policy queries with its predictive world queries, transferring the learned spatial dynamics into policy latents and using them to condition action generation. At deployment, only the policy is retained, operating on RGB observations, language, and robot state without depth input or explicit future prediction. On 11 DexJoCo simulation tasks, DexForesight improves the overall success rate of the base policy from 49.4% to 60.5%. On three real-robot dexterous manipulation tasks, DexForesight achieves 58.3% success, compared with 50.0% for the strongest baseline. Project page https://dexforesight.github.io/
WorldMind: Test-Time Scaling for Visual Spatial Reasoning through Latent World Simulation
Vision-Language Models (VLMs) have continuous deficiency in spatial reasoning tasks that require mental spatial transformation, such as reasoning under hypothetical changes in viewpoint or position, known as perspective-taking tasks. Existing approaches either scale linguistic reasoning or augment VLMs with static geometric representations, but neither directly enables active transformation of the represented scene. World models offer this capability, yet existing systems typically communicate through synthesized RGB images, introducing a latent-to-pixel-to-latent bottleneck. We introduce WorldMind, a framework for test-time scaling of spatial reasoning through latent world simulation. Given a query and observation, the VLM selects useful spatial transformations, while a world model executes them and exposes aligned intermediate latent features directly to the VLM. The resulting latent evidence is incorporated into subsequent reasoning steps, enabling iterative spatial simulation without repeatedly decoding imagined states into pixels. To optimize this interaction, WorldMind further employs process-verifiable reinforcement learning that rewards reasoning progress only when the associated action and latent transition are valid. WorldMind thus scales spatial reasoning by allocating additional inference-time computation to active world simulation rather than generating more language or pixels. Across VSI-Bench, MMSI-Bench, and MindCube-Tiny, WorldMind achieves an average improvement of 16.0% over the base VLM and 8.3% over the strongest RGB-based world simulation.
FloMo: 3D Scene Flow as a World Action Model Intermediate for Learning from Human Video
Pretrained video generation models distill internet-scale motion priors into robot policies, but full video is a redundant signal for action prediction. We argue that dense 3D motion is the medium through which video pretraining best transfers to robot actions, and that predicting motion rather than video affords better sample-efficiency for action prediction. FloMo is a 5B video-generation model fine-tuned to predict 3D scene-flow and action tokens. Rendering scene flow as RGB and encoding it into the same latent space as video enables a pretrained video generation backbone to efficiently adapt to predict 3D scene flow. Co-trained on action-free human video and action-labeled robot teleoperation data, our model benefits from human-video pretraining and generalizes to unseen objects, motion primitives, and tasks. Replacing motion prediction with a video prediction objective degrades out-of-distribution generalization, while adding a video prediction target alongside motion lowers policy success rather than improving it. When motion, video, and action are predicted jointly, action tokens attend far more to motion than to video, suggesting that predicting video wastes model capacity. Together, these results indicate that motion effectively mediates the transfer from video pretraining to action. More information can be found at flomo-wam.github.io.
Trusting Imagined Futures: Auditing Imagined Rewards for VLA Post-Training
Video world models enable candidate actions to be evaluated through imagined outcomes, but high predicted rewards can favor trajectories with poor execution quality. We introduce Trusting Imagined Futures (TIF), an offline supervision selection method for vision-language-action (VLA) post-training that audits predicted rewards using execution feedback. A multidimensional reward model predicts a utility score for candidate ranking and five dimensions of execution quality. An auditor uses reward predictions and uncertainty to identify overestimation patterns learned from paired imagined videos and real executions. TIF accepts the candidate with the highest predicted utility when its advantage over the existing trajectory exceeds a margin and its audit risk passes a threshold; otherwise, it retains the existing supervision. The retained real trajectories then supervise behavior cloning. Across four LIBERO suites and three training seeds, TIF achieves a mean success rate of 76.25%, improving over training on the existing trajectories by 3.00 percentage points with the same trajectory count and training budget. Adding audit gating yields 1.47 percentage points with the reward model and advantage margin held fixed. The gains persist against advantage ranking and logistic regression risk selection at matched replacement counts. On recorded videos from three real robot tasks, TIF increases clean completion among selected trajectories from 58.33% to 70.00% across 60 candidate pools.
HiFOR: Hierarchical Foresight with Action-Guided World Grounding for Vision-Language-Action Model
Vision-Language-Action (VLA) models have demonstrated promising capabilities in robotic manipulation by integrating visual perception, language understanding, and action generation. Recent advances have further incorporated future prediction to enhance the understanding of environmental dynamics and task progression. However, existing methods often model short-term dynamics and long-term task objectives separately, while directly incorporating predicted futures into action policies without sufficiently considering their relevance to current action requirements. To address these limitations, we propose HiFOR, a hierarchical foresight framework with action-guided world retrieval. Specifically, Hierarchical Foresight Prediction jointly predicts short-term state evolution, semantic milestones, and final goal states through a Hierarchical Dependency Mask (HDM), establishing dependencies across temporal horizons. Building upon these predictions, Structure-Aware Latent World Modeling organizes future latent representations along the horizon, view, and spatial dimensions to facilitate fine-grained retrieval. Furthermore, Action-Guided Factorized World Retrieval adaptively retrieves relevant future information from the structured latent world based on evolving action representations during denoising, providing targeted guidance for action generation. Extensive experiments on two simulation benchmarks and six real-world robotic manipulation tasks demonstrate the effectiveness of our approach.
PRIME W0: Physical Priors from World Models for Embodied Manipulation
Vision-Language-Action (VLA) models excel at instruction following, yet most remain feed-forward: they condition actions only on the current observation and discard both long-horizon temporal context and the hierarchical structure of VLM representations. We present PRIME W0, a hierarchical conditional flow-matching framework that closes this gap through three synergistic designs: (1)per-layer KV injection from a pretrained VLM into an Adaptive MoE action expert; (2)dynamics-aware conditioning via asynchronous world-model subgoal-image prediction and a DINOv2 historical-frame semantic bridge; and (3)Action Dynamic Manifold (ADM) flow matching with covariance-structured noise that respects robot kinematic coupling. PRIME W0 is trained in four stages---VLM pre-training, DAT/NTP visuomotor pre-training, DiT flow-matching decoder assembly with WM injection, and end-to-end SFT---and achieves state-of-the-art results on LIBERO and real-robot benchmarks, with systematic ablations validating each architectural component.
Utility-WAM: Correcting Future-Induced Action Changes
World Action Models (WAMs) can condition robot actions on generated visual futures, but this additional context is neither cost-free nor uniformly beneficial. We examine what a predicted future actually changes: the action delta between the future-conditioned and future-free execution paths of the same model. Under a shared WAM checkpoint, this delta barely shifts average benchmark success ( points on LIBERO, on RoboTwin2.0), while generating visual futures increases inference latency by 1.7--1.9. These aggregate metrics conceal a concentrated failure mode that we term future-induced negative transfer: while predictive deltas benefit long-horizon tasks ( points on LIBERO-Long) and remain negligible on most routine manipulations, they induce severe degradation on precision tasks---with Press Stapler and Place Can Basket losing 54.0 and 34.0 points, recurring as top failure modes across independently trained WAM variants. We introduce Utility-WAM for in-place correction of future-conditioned control. Its primary learning component is a residual adapter that repairs future-induced velocity deltas under a frozen shared WAM by aligning them with ground-truth dynamic guidance and penalizing excess error beyond the future-free baseline. The adapter selectively eliminates the majority of severe task drops (reducing losses to 6.0 and 14.0 points) while preserving long-horizon foresight. To eliminate superfluous video computation, a lightweight pre-generation gate reuses counterfactual action outcomes to selectively bypass video generation on 75.5% of RoboTwin queries, cutting latency by 38% while improving average success to 93.3% on RoboTwin2.0 and 98.9% on LIBERO. On physical bimanual manipulation across three real-world tasks, Utility-WAM succeeds in 53 of 60 trials, compared to 43 for FAST-WAM.
ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning
Vision-Language-Action (VLA) models have made substantial progress in robotic manipulation, but their generalization remains constrained by the diversity and coverage of action-labeled robot demonstrations. Human videos provide much broader behavioral and environmental experience, while learning from them faces a fundamental trade-off between scalable data and explicit action grounding. We introduce ThinkWorld, a VLA learning framework that grounds action reasoning in world dynamics and strengthens it through reinforcement learning with world imagination. ThinkWorld represents each action thought as a Chain-of-Dynamics (CoD), a sequence of discrete motion tokens that captures the intermediate dynamics connecting the current world to a future state and serves as a common action representation across embodiments. We first initialize CoD with action-labeled demonstrations, then predict the future induced by each CoD and reinforce those whose predictions better match the observed outcome. This enables videos without action annotations to provide direct supervision from observed state transitions, allowing human and robot experience to refine the same reasoning space. Experiments in simulation and on real robots demonstrate effective human-to-robot transfer and strong generalization to unseen tasks and environments.
PDRVLA: A Vision-Language-Action Model with Physical Dynamics Representations
Vision--Language--Action (VLA) models generate robot actions from visual observations and language instructions, but their understanding of physical dynamics remains limited. Video world models can capture scene dynamics, yet generating future videos introduces high latency into the control loop. We introduce PDRVLA, whose lightweight Physical Dynamics Module extracts physical tokens from vision--language features to condition action generation without online video generation. The module learns through direct prediction of future RGB, depth, and optical-flow features, with auxiliary supervision through a frozen pretrained video world model. Training relies entirely on publicly available data, expanded through human-to-robot synthesis, subtask annotation, and embodied chain-of-thought (ECoT) annotation. Across four simulation benchmarks, PDRVLA achieves success rates of 99.3% on LIBERO, 89.9% on LIBERO-Plus, 93.6% on RoboTwin~2.0, and 81.5% on SimplerEnv. Ablations show improved future prediction with world-model supervision and evaluate how subtask conditioning, future object-pose supervision, and ECoT training affect action success.
Vision-Language Agents Can Learn in Their Own Latent Worlds
Learning from experience is essential for building the next generation of agents that continually expand their capabilities. Pretrained vision-language models (VLMs) already internalize rich world knowledge, yet this knowledge is rarely leveraged to drive policy improvement through online interaction. Realizing this potential efficiently remains challenging, as policy updates shift the agent's interaction distribution, requiring the world model to adapt accordingly, while continually updating large pretrained models is computationally expensive. To address this challenge, we introduce Dwell, a model-based reinforcement learning (RL) framework that reframes a VLM as a recurrent latent transition operator, endowing it with the ability to predict, adjust, and act in latent space. Leveraging the VLM's own representations and its dual role as both world model and policy, the agent continually learns from real interactions to anticipate and evaluate action outcomes, and from latent imagination to improve its behavior. Across five interactive visual environments, Dwell achieves the strongest overall performance among trainable methods while substantially reducing the compute required for imagination, showing that effective world models for VLM agents need not reconstruct the future; they need to predict futures that are useful for action. Our code is publicly available at https://github.com/Tom111-lhx/Dwell
Future Prediction as Training-Only Video Supervision for Pretrained VLA Policies
World action models (WAMs) introduce predictive supervision into robot policy learning by jointly modeling future visual observations and actions. Pretrained vision-language-action (VLA) policies acquire rich visual and semantic representations as well as action-generation capabilities. Their action objectives typically lack direct supervision for predicting future visual observations. Future videos in robot demonstrations record object motion and task progress. They can provide temporal visual supervision that complements action labels. We propose Training-Only Video Supervision, which introduces future-video prediction as an auxiliary objective for VLA fine-tuning. We condition a randomly initialized video-prediction expert on the VLA's shared vision-language features. The expert learns to predict future video latents from robot demonstrations. The video prediction loss and the original action loss jointly update the shared vision-language backbone, providing additional temporal visual supervision. After training, we remove the video expert and its adapter. The fine-tuned VLA retains its original action-generation pathway at inference. On the 50-task RoboTwin 2.0 benchmark, our method achieves 71.18% average success, compared with of 58.33%. In a matched evaluation on four real-world manipulation tasks, success increases from 42.50% to 61.25%. These results show that future visual modeling can serve as auxiliary training supervision for pretrained VLAs. The fine-tuned action policy benefits from this supervision without retaining a video-prediction branch at deployment.
-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model
Humans reason about the physical world by flexibly combining different forms of inference as a problem unfolds. Achieving this flexibility is a major open challenge in AI: capabilities often remain separated across specialized models, limiting how they can be composed and jointly adapted to a task. We introduce -0.6, a unified vision-language world model that adaptively composes perception, language, geometry, and simulation within a shared reasoning process we call streams of thought. These streams interleave images, text, and explicit representations of geometry and motion, all of which the model can generate and interpret through a shared token interface. -0.6 surpasses leading open-weight models zero-shot on multi-view spatial reasoning, with further gains from reasoning through geometric intermediates. It also supports counterfactual simulation: generating and interpreting imagined scenes to determine, for example, whether moving an object would obstruct a pathway. Reinforcement learning from task feedback jointly optimizes the selection, generation, and use of intermediate representations, improving visual question-answering accuracy with gains that transfer to unseen benchmarks. -0.6 offers a path toward adaptive physical reasoning -- not only predicting the physical world but also choosing the representations through which to reason.
See What I Need, Dream What I Want: A Navigation Foundation Model on a Self-Constructed Representation Space
What visual representation should a navigation policy act on? Navigation policies built on vision foundation models largely leave representation design to the encoders' pretraining objectives. We present RepresentNav, a navigation foundation model built on a self-constructed representation space: a visual representation constructed by the policy's own objectives from vision foundation models. RepresentNav builds this space with Representation Residuals, a compact three-layer transformer that attends over features spanning the full depth of multiple complementary vision foundation models, so that the policy learns to see what it needs for navigation. In the same space it dreams what it wants: before every action a latent chain-of-thought predicts the representation of the scene a moment ahead. The dream is trained end-to-end with the policy as a joint-embedding predictive architecture (JEPA), with a next-representation prediction loss and a regularizer on its targets. RepresentNav sets a new state of the art on VLNVerse, VLN-CE R2R, EVT-Bench, and two point-goal benchmarks, and its gains carry over to a physical robot amid clutter and moving people. On VLNVerse fine-grained, RepresentNav achieves 83.80% success, outperforming the previous state of the art by 20.05 percentage points, and on EVT-Bench distracted tracking it reaches 85.5% with a single camera (+10.1). With the same encoders, the self-constructed space clearly outperforms concatenating their outputs and learns to weight their layers unevenly, and a dream in it cuts collisions by a quarter. A single dreamed latent decodes into the depth, semantics, and 3D occupancy of the scene a moment ahead, more accurately than pretrained encoders applied to the real future frame, although occupancy is never supervised, and the rollout tracks the robot's own motion over the next second: the dream emerges as a compact latent world model.
DRILL: Training World Models to Improve Policies, Not Predict Pixels
World models are increasingly used to reduce environment interaction when post-training vision-language-action (VLA) policies with reinforcement learning. Yet they are typically trained to predict pixels one step ahead, even though their role here is not merely to render visually faithful videos but to produce learning signals from imagined rollouts that improve the policy. We show that this mismatch is substantial: lower pixel MSE can produce worse policy gradients, and a world model trained for policy utility can yield better downstream policies despite 17% higher pixel MSE on the next frame. We propose the Downstream-Return Imagination Learning Loop (DRILL), a bilevel framework that updates the world model to maximize the policy's return after an inner GRPO update on imagined rollouts. DRILL has two instantiations. DRILL-VWI is a closed-form surrogate, requiring no additional simulator rollouts, that reweights prediction errors by policy visitation, the group-relative GRPO advantage, and model confidence. DRILL-IMG is a full meta-gradient method that differentiates through the inner update using truncated Hessian-vector products. We show that DRILL-VWI recovers the local one-step component of the DRILL-IMG meta-gradient under score-compatibility and small-residual assumptions. Across five manipulation simulators and two VLA backbones, DRILL-IMG raises mean success rate by 13.1 points over WMPO and 7.9 points over RLVR-World, and complements RLVR pretraining rather than competing with it. More broadly, we believe video world models for VLA reinforcement learning can be productively trained by the policy gradients they produce, not only by pixel-level prediction.