CAG-WAM: Causal Action-Gain World Action Models for Robot Manipulation

Peizheng Guo; Cheng Zhang; Xingyu Zhang; Changwen Zheng; Wenwen Qiang

cs.ROOct 10, 2026

CAG-WAM: Causal Action-Gain World Action Models for Robot Manipulation

Authors: Peizheng Guo, Cheng Zhang, Xingyu Zhang, Changwen Zheng, Wenwen Qiang

Organizations: National Key Laboratory of Space Integrated Information System, Institute of Software Chinese Academy of Sciences, Beijing, China · University of Chinese Academy of Sciences, Beijing, China · Roban

Abstract

World Action Models (WAMs) combine action generation with future prediction, allowing predicted futures to support action evaluation and robot decision making. However, the future associated with a candidate action may contain action-induced and environment-driven changes. Directly evaluating the predicted future may not fully characterize the future change associated with the candidate action. For example, a candidate action that worsens a favorable environmental future may still yield a favorable outcome, while one that improves an unfavorable environmental future may still yield an unfavorable outcome. To analyze this issue, we formulate a Structural Causal Model (SCM) for action evaluation in WAMs. The SCM formalizes how action-induced and environment-driven changes jointly shape future evaluation, and reveals a backdoor path through which environmental evolution can also affect the future evaluation associated with a candidate action. Intervention and backdoor adjustment account for this path, providing a causal basis for characterizing the future change associated with each candidate action. Based on this analysis, we propose Causal Action-Gain World Action Model (CAG-WAM). For each candidate action, CAG-WAM predicts an action-conditioned future and an environmental future under the same current observation, and defines their evaluation difference as causal action gain (CAG) for characterizing the future change associated with the candidate action. During training, we incorporate CAG into Group Relative Policy Optimization (GRPO) to guide robot policy learning. During inference, we evaluate the initial action with CAG and, when needed, revise it using predicted future information. Experiments on multiple benchmarks support the effectiveness of CAG-WAM.

Figures & tables

Figure 1: Motivation and causal analysis. (a) Illustrative cases where the overall future evaluation may not fully reflect the future change associated with a candidate action. (b) Structural causal model of action evaluation, including the backdoor path $\mathbf{A}_{t}\leftarrow\mathbf{O}_{t}\rightarrow\mathbf{E}_{t+1}\rightarrow\mathbf{O}_{t+1}\rightarrow Y_{t}$ .

Figure 2: CAG-WAM Architecture. (a) Future-Conditioned Action Model generates action chunks from the current observation and task instruction, and can additionally incorporate future feedback for action revision. (b) Paired-Future World Model predicts action-conditioned and environmental futures under different action conditions and evaluates them using the same future evaluator.

Figure 3: Training and inference framework of CAG-WAM. (a) Training: candidate actions are evaluated by comparing their action-conditioned futures with the environmental future, and the resulting causal action gain is used as the policy-optimization advantage. (b) Inference: the initial action is evaluated using the same paired-future comparison; when its causal action gain falls below the revision threshold, future feedback is used for an additional action generation before execution.

Method	Pick&Place	Door	Drawer	Knob	Lever	Button	Insertion	Avg.
3D Diffuser Actor ( Ke et al., 2024 )	0.0	3.5	0.0	6.0	20.0	15.3	0.0	5.5
DP3 ( Ze et al., 2024 )	1.5	40.5	42.0	15.0	42.7	36.0	11.0	22.8
Diffusion Policy (ResNet) ( Chi et al., 2025 )	6.2	67.0	73.0	27.0	65.3	68.7	27.0	40.6
Diffusion Policy (CLIP) ( Chi et al., 2025 )	6.0	61.5	78.0	18.0	54.0	85.3	36.0	40.7
GR00T N1 ( Bjorck et al., 2025 )	22.6	58.0	89.0	41.5	63.0	78.0	48.0	49.7
FPV ( Donat et al., 2025 )	14.2	82.0	83.0	43.0	66.7	84.7	32.0	50.5

Table 1: Success rates (%) on the 24-task RoboCasa benchmark. The tasks are grouped into seven manipulation skill families: Pick&Place (8 tasks), Door (4 tasks), Drawer (2 tasks), Knob (2 tasks), Lever (3 tasks), Button (3 tasks), and Insertion (2 tasks). Avg. denotes the average success rate over all 24 tasks. Full results are provided in the Appendix .

Method	Spatial	Object	Goal	Long	Avg.
Diffusion Policy ( Chi et al., 2025 )	78.3	92.5	68.3	50.5	72.4
Octo ( Team et al., 2024 )	78.9	85.7	84.6	51.1	75.1
OpenVLA ( Kim et al., 2024 )	84.7	88.4	79.2	53.7	76.5
TraceVLA ( Zheng et al., 2025 )	84.6	85.2	75.1	54.1	74.8
SpatialVLA ( Qu et al., 2025 )	88.2	89.9	78.6	55.5	78.1
UniVLA ( Bu et al., 2025 )	96.5	96.8	95.6	92.0	95.2

Table 2: Success rates (%) on the four LIBERO task suites.

Action-Conditioned World Models

Action-conditioned world models predict how environments change under actions to support planning, control, and simulation.

See all 346 papers

Sep 13, 2026

World-Action Models for Robot Learning and Control: A Survey

Robots operating in open environments act under partial observability, physical constraints, and dynamic task contexts. Beyond mapping observations and language instructions to actions, they must anticipate how candidate actions may affect future states and task-relevant outcomes. Recent advances in world models, video generation, and Vision-Language-Action (VLA) policies have motivated the development of World-Action Models (WAMs), which couple future world prediction with executable action generation. This survey provides a robotics-oriented review of WAMs. We clarify their scope relative to conventional world models, model-based reinforcement learning, action-conditioned video generation, and reactive VLA policies, and organize existing methods through a unified taxonomy covering representations, transition modeling, action interfaces, architectures, training pipelines, data modalities, and scaling strategies. We further review applications of WAMs in manipulation, navigation, and autonomous driving, and we summarize the datasets, benchmarks, metrics, and protocols used to evaluate WAM systems. Finally, we discuss key challenges in action alignment, world-action factorization, spatial and multi-view consistency, long-horizon memory, neural simulation for closed-loop policy learning, and efficient inference. Taken together, this survey aims to provide a concise technical foundation for integrating predictive world modeling with action generation, toward more reliable embodied robot intelligence. Project page: https://rcl-robotics.github.io/Awesome-World-Action-Models.

Zuxing Lu, Hongjia Zhai, Guanzhi Wang +13

MBZUAI · Caltech · Amazon FAR +4

World Action Models Action-Conditioned World Models World Models for Robotics World Model Learning World Model-Based Planning

Ask PDF

Oct 29, 2026

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

Δa

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 (

+0.8

points on LIBERO,

-0.8

on RoboTwin2.0), while generating visual futures increases inference latency by 1.7--1.9 $\times$ . These aggregate metrics conceal a concentrated failure mode that we term future-induced negative transfer: while predictive deltas benefit long-horizon tasks ( $+2.4$ 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.

Feihong Zhang, Xi Lin, Yulong Shi +7

Tsinghua University · Johns Hopkins University · Northeastern University +4

World Action Models World Models for VLA Models Efficient Inference for World Action Models

Ask PDF

Aug 1, 2026

SelfWAM: A Self-Grounded Unified World Action Model for Fast Robot Control

World Action Models (WAMs) improve robot policy learning by jointly modeling actions and future observations. However, conditioning future prediction only on the task prompt and observation context risks capturing generic task progression rather than the action-specific consequences of the executed action. We introduce SelfWAM, a unified self-grounded WAM built on a modality-specialized Mixture-of-Transformers (MoT) architecture that jointly predicts actions, action-conditioned future RGB frames, and robot self-masks, thereby grounding future prediction in the robot's visible body and its action-induced motion. During joint training, SelfWAM allows future visual queries to attend to a clean copy of the demonstrated action, turning the video branch into an action-specific consequence model while leaving the fast action-only inference path unchanged. To focus video learning on action-relevant visual changes, we use prompt-specific objectives for future robot self-mask prediction, which removes appearance details and provides a target whose temporal evolution is tightly coupled with the conditioning action. Together, clean-action conditioning and future self-mask supervision make future predictions more directly reflect how the executed action changes the robot's visible motion and the surrounding scene. Experiments on RoboTwin 2.0 and real-world manipulation tasks show that SelfWAM produces more action-sensitive futures and preserves fast policy inference, while improving policy performance.

Bikang Pan, Fan Liu, Haotao Lu +2

1ShanghaiTech University · 2InstAdapt

Action-Conditioned Video Generation World Models for Robotics Visuomotor Policy Learning Robotic Manipulation World Action Models

Ask PDF

Abstract

Action-Conditioned World Models

Explore similar work

World-Action Models for Robot Learning and Control: A Survey

Utility-WAM: Correcting Future-Induced Action Changes

SelfWAM: A Self-Grounded Unified World Action Model for Fast Robot Control