Acting Through Registers

Yiwei Zhang; Yantai Yang; Jin Gao; Jin Lou; Andong Chen; Yufei Liu; Boyang Xing; Yuchen Zhu

cs.ROOct 13, 2026

Acting Through Registers

Authors: Yiwei Zhang, Yantai Yang, Jin Gao, Jin Lou, Andong Chen, Yufei Liu, Boyang Xing, Yuchen Zhu,

Organizations: AutoLab, School of Artificial Intelligence, Shanghai Jiao Tong University · Institute of Automation, Chinese Academy of Sciences · Humanoid Robot (Shanghai) Co., Ltd.

Abstract

World Action Models (WAMs) use generative visual representations to condition robot actions, yet how these representations organize action-relevant information remains poorly understood. We identify persistent high-norm spatial outliers in WAM representations and find that their action relevance persists even after normalization and K/V projection attenuate their extreme magnitude. Motivated by this observation, we introduce RegisterWAM, which adds learnable non-spatial registers to the world model and uses them directly for action conditioning. Registers alone do not fully eliminate spatial outliers; pairing them with observation flow-matching (FM) loss supervision suppresses spatial outliers while concentrating high-norm responses in registers. Beyond this reorganization, observation FM loss supervision also improves action learning without registers, motivating an additional noised-observation FM objective that learns current-scene recovery from corrupted inputs. The resulting model achieves 81.64% success on LIBERO-Plus without additional embodied pretraining data or foundation models, reaches 93.0% average success on RoboTwin 2.0, and improves real-world manipulation performance.

Figures & tables

Panel of Figure 1 — Figure 1: Spatial outliers and action relevance. Hidden-state norms and masked-token action sensitivity across video models and FastWAM.

Figure 2: Layer-wise spatial outlier strength. High-norm outliers are more pronounced and consistent in the examined FastWAM checkpoint.

Panel of Figure 3 — Figure 3: From hidden states to action. The magnitude anomaly vanishes after K/V projection, while its action relevance persists.

Figure 4: Action-sensitivity ranks. Dots mark the outlier token.

Panel of Figure 5 — Figure 5: Output-layer representations with registers. (a) Without observation FM loss supervision, spatial outliers remain alongside high-norm registers. (b) With observation FM loss supervision, spatial-token norms are regularized while high-norm responses remain concentrated in registers.

Figure 6: Register interfaces. Curves compare Only Reg. and Raw + Reg. conditioning with observation FM loss supervision. Raw denotes dense observation tokens only. Dashed lines show FastWAM, raw plus 8 registers without observation FM loss supervision, and raw tokens with observation FM loss supervision but no registers.

Figure 7: RegisterWAM architecture. Registers condition the Action DiT; visual outputs receive FM loss supervision. Encoded robot-state tokens are appended to the text embeddings (omitted for clarity).

Figure 8: Training attention mask. Action tokens attend to registers and themselves. Gray entries are masked.

Method	Emb. PT.	Extra Prior	Camera	Robot	Lang.	Light	Backg.	Noise	Layout	Avg.
$\pi_{0.5}$ [ 4 ]	$\checkmark$	–	75.4	77.5	85.6	96.9	94.6	89.7	85.7	85.7
FastWAM [ 38 ]	–	–	16.4	44.5	68.9	78.2	53.7	37.7	60.7	51.5
Faster-WAM [ 22 ]	–	–	67.9	49.0	92.1	94.3	57.0	82.3	82.7	75.0
Faster-WAM [ 44 ]	–	–	53.8	71.6	94.7	96.3	61.3	63.6	79.1	73.6
4D-WAM [ 34 ]	–	$\checkmark$	45.2	64.3	90.6	94.3	57.7	69.1	79.2	71.0
ST-WAM [ 30 ]	–	$\checkmark$	55.4	60.1	79.3	93.0	74.2	79.5	74.3	72.8

Table 1: LIBERO-Plus success rates (%). Emb. PT.: embodied-policy pretraining; Extra Prior: external pretrained models, modalities, or geometric/semantic/motion supervision.

Table 10

Method	Batteries	Stacking	Handover	Height	Tablecloth	Distractor	Avg.
	In-distribution			Out-of-distribution
FastWAM	0.5000	0.0000	0.6200	0.5500	0.0000	0.4200	0.3483
RegisterWAM	0.6875	0.2500	0.7550	0.7300	0.0000	0.7200	0.5238

Table 4: Real-world task scores (20 trials per setting). OOD variants modify the bottle-handover task; Avg. equally weights all six settings.

Panel of Figure 9 — Figure 9: Scenes for the three real-world ID tasks. The three OOD variants modify the bottle-handover setting.

Table 13

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 10: Wan2.2 first-frame outliers versus future noise. Color denotes first-frame spatial outlier strength $S_{\mathrm{out}}$ . Pronounced peaks occur mainly in middle-to-late layers under strong future noise.

Panel of Figure 11 — Figure 11: Additional FastWAM outlier visualizations. Hidden-state norms for three observations in FastWAM and two retrained variants. All panels show norms overlaid on the corresponding observations. Colors should be compared within each panel.

Figure 12: Complementary effects of registers and observation FM loss supervision. Layer-wise spatial outlier strength $S_{\mathrm{out}}$ across ten observations. Registers suppress intermediate-layer peaks, while observation FM loss supervision suppresses outliers in the final layers; together, they suppress pronounced peaks throughout the network.

Panel of Figure 13 — Figure 13: RegisterWAM spatial norm maps. Hidden-state norms overlaid on two observations, without conspicuous localized outliers. Colors are interpreted within each panel.

Panel of Figure 14 — Figure 14: Register attention over spatial tokens. Four selected register–head pairs, overlaid on observation 003426. Register and head indices are shown below each panel.

Action-Conditioned World Models

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

See all 346 papers

World-Action Models for Robot Learning and Control: A Survey
Sep 13, 2026Zuxing Lu, Hongjia Zhai, Guanzhi Wang +13World Action Models Action-Conditioned World Models
MBZUAI · Caltech · Amazon FAR +4
RepWAM: World Action Modeling with Representation Visual-Action Tokenizers
Jun 11, 2026Junke Wang, Qihang Zhang, Shuai Yang +5Latent Action Learning World Models for Robotics
Institute of Trustworthy Embodied AI, Fudan University · 2Robbyant, Ant Group · 3Hongkong University of Science and Technology
Copper-Policy: Focus on the Representation for Robust Robot Manipulation
Sep 26, 2026Zexin Feng, Yixu Feng, Lingyu Xiao +6Robot Policy Learning Robotic Manipulation
The University of Hong Kong · The University of Sydney · Tsinghua University +1

Sep 13, 2026J/K move · Enter open · S save

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

Zuxing Lu, Hongjia Zhai, Guanzhi Wang, Huajian Zeng+12

MBZUAI · Caltech · Amazon FAR+4

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

Open paper AskPDF arXiv

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.

Abstract

Action-Conditioned World Models

Explore similar work

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

RepWAM: World Action Modeling with Representation Visual-Action Tokenizers

Copper-Policy: Focus on the Representation for Robust Robot Manipulation