ActionSplice: In-Flight Action Editing for Interactive World Models

Pardis Taghavi; Tingyu Guo; Jonas Lossner; Gaurav Pandey; Reza Langari

September, 2026World Models

ActionSplice: In-Flight Action Editingfor Interactive World Models

Pardis Taghavi, Tingyu Guo, Jonas Lossner, Gaurav Pandey, Reza Langari

Texas A&M University

arXiv Code

TL;DR

ActionSplice edits an in-progress solver state when the requested action changes during sampling. CST-R retargets the active chunk; CST-T preserves a temporal prefix at the intervention step and corrects only the suffix. Both resume the frozen world model from the current solver step without replaying completed evaluations.

Method

Same-step state transport.

When an action changes after solver evaluation r, ActionSplice corrects the interrupted backbone-native solver state toward its matched counterfactual at the same step. Sampling then continues from the next evaluation—without replaying completed world-model evaluations.

ẑ_r=z_r⁻+M_m⊙C_{φ_q}(𝓘_r, M_m)

(1 − M_m)⊙(ẑ_r − z_r⁻)=0

CST-R: m = 0 · retarget the entire active chunk.
CST-T: 0 < m < T · preserve the old-action prefix and edit the suffix.
Inference: Resume the remaining K − r evaluations with the backbone, sampler, and history frozen.

Full rollback provides training targets and evaluation oracles; it is not executed during ActionSplice inference.

Qualitative

CST-R applies the revised action within the active chunk.

Backward → Yaw left · r = 2

Backward → Forward · r = 2

Forward → Yaw left · r = 2

Repeated updates

CST-R handles repeated in-flight action updates.

Matched HY-WM1.5 rollouts with repeated action interruptions at r = 2.

Five sequential action updates · r = 2

Two sequential action updates · r = 2

CST-T / Qualitative

CST-T: Within-chunk action transport

Matched Wait controls versus CST-T · r = 2.

Results

Evaluation highlights.

Quantitative results reported in the current manuscript draft.

CST-R · rollback fidelity

61.5%lower LPIPS on minWM

75.9%lower LPIPS on HY-WM1.5

Relative to direct condition swapping.

CST-T · suffix fidelity

56.1%lower LPIPS on minWM

77.5%lower LPIPS on HY-WM1.5

Relative to direct condition swapping.

CST-R · pixel-ready speedup

2.69×minWM

1.64×HY-WM1.5

Relative to waiting.

HY-WorldPlay · CST-R

25.66PSNR

0.6902SSIM

0.1337LPIPS

Against the original rollout.

Citation

@article{taghavi2026actionsplice,
  title={ActionSplice: In-Flight Action Editing for Interactive World Models},
  author={Taghavi, Pardis and Guo, Tingyu and Lossner, Jonas and Pandey, Gaurav and Langari, Reza},
  journal={arXiv preprint arXiv:2609.08230},
  year={2026}
}