Skip to content
AI News HubLIVE
Original source2 min read

Refinement vs. Replanning: Inference Compute Allocation in Robot Policies

Summary

This paper studies how a frozen chunked robot policy can allocate deployment-time inference compute between refining action chunks within each policy call and replanning more frequently from fresh observations. Using a SmolVLA checkpoint in LIBERO simulation, the authors find that with h=2, d=2 outperforms d=5 (139/160 vs 122/160) while using less arithmetic, and that h=5 vs h=2 shows similar success but 60.7% lower arithmetic per attempt. The work emphasizes joint evaluation of generation effort and feedback frequency without defining a universal setting.

SourcearXiv RoboticsAuthor: Alba Maria Tellez Fernandez
Refinement vs. Replanning: Inference Compute Allocation in Robot Policies
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 14 Nov 2026]

Title:Refinement vs. Replanning: Inference Compute Allocation in Robot Policies

View a PDF of the paper titled Refinement vs. Replanning: Inference Compute Allocation in Robot Policies, by Alba Maria Tellez Fernandez

View PDF HTML (experimental)

Abstract:A frozen chunked robot policy can spend deployment-time inference compute in two coupled ways: more work within each policy call to refine a predicted action chunk, or more frequent policy calls to replan from fresh observations. We study this allocation without updating model weights. Our evaluation procedure reports success with both nominal compute rate and complete attempt cost, selects comparisons on discovery tasks, and tests prespecified comparisons on held-out tasks. Comparisons added after inspecting the held-out results only summarize observed patterns. We demonstrate the procedure with one frozen SmolVLA checkpoint in synchronous LIBERO simulation, using denoising count as the refinement control and fixing chunk length at K=50. Arithmetic and latency characterize cost differently: at d=10, 82.1% of counted arithmetic is independent of denoising, whereas a descriptive affine fit to fixed-input host time assigns 17.7% to its intercept. At h=2, d=2 succeeds on 139/160 attempts versus 122/160 for d=5 (+10.6 percentage points; task-conditional 95% interval: +4.4 to +16.9), while using 5.9% less arithmetic per control step. Executing h=5 rather than h=2 at d=2 yields 138/160 versus 139/160 successes and 60.7% lower arithmetic per attempt; equivalence is not established. The case study shows why deployment adaptation must evaluate generation effort and feedback frequency jointly; it does not define a universal setting.

Comments: NeurIPS 2026 Workshop: Post-Training Adaptation of Robot Foundation Models

Subjects:

Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)

Cite as: arXiv:2611.19034 [cs.RO]

(or arXiv:2611.19034v1 [cs.RO] for this version)

https://doi.org/10.48550/arXiv.2611.19034

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Alba María Téllez Fernández [view email] [v1] Sat, 14 Nov 2026 00:16:54 UTC (1,229 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Refinement vs. Replanning: Inference Compute Allocation in Robot Policies, by Alba Maria Tellez Fernandez

View PDF

HTML (experimental)

TeX Source

view license

Additional Features

Audio Summary

Current browse context:

cs.RO

new | recent | 2026-11

Change to browse by:

cs cs.AI cs.LG

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

InvestorsAdvanced

Key points

  • Studies joint allocation of refinement compute and replanning frequency for frozen robot policies
  • In LIBERO simulation, d=2 with h=2 achieves higher success (139/160) than d=5 (122/160) with 5.9% less arithmetic per control step
  • Arithmetic and latency characterize cost differently, requiring both to be reported
  • Evaluation procedure uses discovery and held-out tasks with prespecified comparisons

Highlights and analysis are generated automatically and may contain errors. Check the original source.