Robot policy learns when and where to stop imagining the future

A preprint adds a learned controller to a robot model that predicts future video before it acts. The controller cuts thinking time by about a third against its closest rival, with small gains in success.

5 min readIIBenchmarkedPreprint, not peer reviewed
Save
A tiny wooden block sits next to fanning sheets of tracing paper with sketched predictions, stopped by a blue thread.AI illustration
Illustration: AI generated

What the study shows

The work concerns world-action models, robot policies that predict future camera images and use that prediction to pick their next movements. They build the prediction by denoising, which means starting from noise and cleaning it up over several steps. Each step costs time. Earlier methods use a fixed number of steps for every situation.

The authors first checked whether a fixed number makes sense. On 10 tasks from the LIBERO-Plus benchmark, the best number of steps ranged from one to four. It also changed between moments within the same task. And the action part of the model paid uneven attention to different regions of the predicted image.

Their method, Elastic-WAM, adds a small learned controller on top of an existing model that stays frozen, so its weights are not changed. The controller makes three choices at each step: whether to stop imagining, how large the next cleaning step should be, and which image regions to refresh. By default it refreshes 96 of the 196 small image pieces, called tokens, and reuses the rest from memory. It is trained in three stages by checking which choices lead to more accurate robot actions.

The authors report results on four public simulation benchmarks. On RoboCasa-GR1, 24 simulated tabletop tasks, success was 57.4% against 55.0% for Joint-WAM, the model it builds on. Time per decision fell by 32.2%, from about 573 to 388 milliseconds. Fast-WAM, which skips video prediction at run time, was quicker still at 331 milliseconds, with 54.9% success.

The largest gain came on LIBERO-Plus, which tests robustness to changes such as camera angle, lighting and sensor noise. Elastic-WAM scored 73.7% against 66.3% for Joint-WAM. On the standard LIBERO and on LIBERO-Pro, the gains were one or two points or less.

On a real two-armed robot, the team ran a four-stage tea-making task 20 times per setting. Elastic-WAM finished the whole task in half the runs under normal and changed lighting, against 30% for Joint-WAM and none for Fast-WAM. With a different cup, both Elastic-WAM and Joint-WAM finished 30% of runs. This is a preprint and has not been peer reviewed.

Our reading

In our reading, the main contribution is efficiency, not accuracy. The simulation gains over Joint-WAM are mostly small, and no spread across repeated training runs is reported. The time saving is larger and easier to trust, because it follows directly from doing less work. A third less time per decision on the same model is a useful result in its own right.

The ablations, tests that remove one part at a time, support the design. Choosing regions with the learned controller beat random choice with the same budget by about four points on RoboCasa-GR1. Refreshing every token did worse than refreshing 96. This suggests that much of the imagined video is not needed for choosing actions, which is a practical insight for anyone building these systems.

We rate this work at level II, Benchmarked. It reports results on public benchmarks against named rivals. The paper names a project page but does not state that code or weights are released, so it does not yet meet level III.

Free membership

Read the rest of this story

Membership is free. You get every article in full and the Issue every Sunday.

We send a sign-in link, no password needed. Already a member? Sign in

Readiness assessment

II
Benchmarked
LevelStatusEvidence
I  Lab onlyMetMethod is described in full, with equations and a training procedure.
II  BenchmarkedMetResults on public benchmarks: LIBERO, LIBERO-Pro, LIBERO-Plus and RoboCasa-GR1, with comparisons against Joint-WAM, Fast-WAM, pi0.5 and others, plus ablations and a small real-world test.
III  Open sourceNot yetThe paper gives a project page (https://lihaitian.com/Elastic-WAM) but states no released code, weights or data. Not confirmed from the paper text.
IV  In productionNot yetNo documented use in production by a named organisation. Real-world test is a lab tea-preparation task.

What would move it up
Releasing code and trained weights, with the project page confirmed to host them, would reach level 3. Level 4 needs documented use by a named organisation outside the authors' lab. Repeated runs with variance and larger real-world trials would also strengthen the evidence.

Checked automatically every week, last on 16 November.

The study

Haitian Li, Tianfu Li, Xuxi Zhao et al.. Elastic-WAM: Adaptive Spatiotemporal Imagination for Efficient Robot Control. arXiv, 10 November 2026. doi:10.48550/arxiv.2611.14257

Preprint, not peer reviewed · Licence arXiv (no reuse) · National University of Singapore, Singapore · University of Hong Kong, Pok Fu Lam · Zhejiang University, Hangzhou

Notes

  1. Li, H., Li, T. et al. Elastic-WAM: Adaptive Spatiotemporal Imagination for Efficient Robot Control. arXiv preprint 2611.14257, 10 November 2026. Not peer reviewed. Source
  2. Project page named in the paper. The paper does not state whether code or weights are hosted there. Source
  3. Exact figures, RoboCasa-GR1 (Table 4): Elastic-WAM 57.4% success at 388.2 ms; Joint-WAM 55.0% at 572.6 ms; Fast-WAM 54.9% at 331.1 ms. Latency on one NVIDIA H800 GPU.
  4. Exact figures, average success: LIBERO 98.8% against 98.5% (Joint-WAM); LIBERO-Pro 54.1% against 52.8%; LIBERO-Plus 73.7% against 66.3% (Joint-WAM) and 69.5% (LingBot-VA).
  5. Ablations: RoboCasa-GR1 learned region choice 57.42%, random 53.83%, all tokens with budget 128 55.17%. LIBERO-Plus: spatial choice only 71.71%, full adaptive 73.71%, budget 196 tokens 70.92%.
  6. Real-world tea task (Table 7), 20 trials per condition on two UFACTORY xArm 7 arms. Full-task success, nominal / lighting / object change: Elastic-WAM 50 / 50 / 30%; Joint-WAM 30 / 30 / 30%; Fast-WAM 0 / 0 / 0%. Progress score with object change: 47.5% against 35.0%.

Edited and signed by

Stefan PerdalEditor and publisher86 stories edited

Stefan read the study, set the readiness level and approved this story. An AI system found the paper and wrote the first draft.

How this was made

An AI model read the study and wrote this article in the structure Science Underway uses. An editor reviewed it and approved it for publication. Check the original study before you rely on the details.

More on Robotics