- Rating
- 2(−1)
- Confidence
- 4(+1)
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper studies whether randomness assigned to discarded action positions can influence the prefix executed by a receding-horizon robot controller. It introduces tail-resampling sensitivity (TRS) to measure this dependence and proposes CommitFlow, which constrains action-token dependencies to make the executed prefix invariant to tail-local noise. A strict causal variant protects every prefix, while a known-boundary variant preserves bidirectional interaction within the committed block. Shared stochastic context is added to support coordination. The paper proves prefix invariance under the stated closure assumptions and demonstrates that tail-only resampling can change closed-loop outcomes in the baseline. The full model improves aggregate success on LIBERO, RoboTwin, and seven AgiBot G1 tasks, although the experiments do not isolate tail invariance as the sole source of these gains.
Critical strengths
- The paper gives a precise and testable formulation of discarded-tail dependence. The dependency constraints address both direct attention paths and indirect paths through conditioning tokens, and the prefix-invariance argument is sound under the stated assumptions. The controlled rollout intervention also shows that this dependence can affect task outcomes, rather than merely changing an action-space diagnostic.
- The evaluation goes beyond a structural zero-TRS check. The matched factorial study supports a positive interaction between prefix closure and shared context, and the pooled LIBERO gain has a paired task-bootstrap interval excluding zero. Longer-horizon simulation and physical trials provide additional evidence, while negative results and action-variation trade-offs are reported explicitly.
Critical weaknesses
- The evidence for dependence is stronger than the evidence that this dependence should be removed. Tail resampling establishes that different random inputs can produce different outcomes, but does not by itself show that tail-local randomness is detrimental. Strict closure with IID noise slightly lowers the success point estimate, and gains remain when no tail is discarded. These findings suggest that the full model benefits from a combination of architectural changes, while the performance value of invariance itself remains less clear.
- The most relevant known-boundary control is not directly matched against the main configurations. Block-5 with IID noise satisfies the required invariance and has a higher observed success rate than the corresponding strict configuration, but the cross-architecture comparison lacks paired task-level contrasts. In addition, the shared-context component combines a correlated prior with an extra conditioning token, leaving their individual contributions unresolved. These gaps limit the justification for the full design.
Questions for authors
- Could you provide a matched comparison of Base, strict closure, and Block-5 at , with paired task-level differences and uncertainty? This would help determine when protecting every prefix is preferable to protecting only the known execution block.
- Can the contributions of the correlated noise prior and the shared conditioning token be separated? Does either component alone recover the success lost under strict closure?
- What are the main failure modes behind the large open_microwave regressions in Table F.2? Do the existing trajectories help explain the relationship between reduced cross-plan revision and increased within-plan variation?
- Could you provide the training and sampling configurations, action normalization and dimensional mappings, and task-level result vectors needed to reproduce the main comparisons?
5 comments
Response to Reviewer MmNm — Part 2/2
A3: Analysis of open_microwave
We are further analyzing the additional open_microwave results and trajectories, focusing on failure patterns and the relationship between cross-plan revision and within-plan action variation. We will provide a separate follow-up on this point.
A4: Reproducibility and per-task results
The newly trained controls use 30,000 updates, global batch size 256, and seed 6113. These LIBERO comparisons use ten Euler steps with and or .
Our LIBERO interface uses an eight-dimensional state: end-effector position (3), axis-angle orientation (3), and gripper joint positions (2). After inverse normalization using the saved training statistics and the configured output transformation, the first seven coordinates of the 32-dimensional model output provide translation deltas (3), rotation deltas (3), and the gripper command (1); the remaining coordinates are padding.
The per-task results and additional configuration details are listed in our separate comment, Per-task LIBERO results and evaluation protocol.
A5: Tail dependence and task performance
The new controls clarify the contribution of shared context to task performance. With the strict mask fixed, the prior-only and token-only variants improve the measured success rate over Strict-IID by 0.50 and 0.75 percentage points, respectively (A2). Prefix closure supplies the invariance guarantee, and shared context supports performance under that constraint.
In the original intervention, the initial state, prefix-local randomness, and all later-call randomness are fixed, while only the first call's tail-local noise changes. Resampling the full discarded tail produces both success and failure in 6 of 40 fixed states for Base, compared with 0 of 40 for CF. This connects the measured dependence to variation in actual task outcomes.
At , Base and Full CF obtain 97.90% and 98.30%, with a difference of +0.40 pp [−0.15, 0.95]. This pattern is consistent with masking and shared context contributing to task performance when the entire chunk is executed.
We will incorporate these experiments and clarifications into the revised manuscript. The supplementary material will include the configuration details and per-task success counts and episode totals for the and comparisons. Thank you again for your constructive feedback. We would be happy to discuss any further questions during the discussion period.
Per-task LIBERO results and evaluation protocol
Supplement to A4: Task-level LIBERO results and configuration
The tables give the supplementary results discussed in A1, A2, and A5. Each numerical entry in a task row is the number of successful episodes out of 50, using official initial-state indices 0-49. The final row reports overall success rates in percent.
Task IDs follow LIBERO's zero-based task order. S/O/G/L denote libero_spatial/libero_object/libero_goal/libero_10, respectively; e.g., S00 is task 0 in libero_spatial.
Method columns: B = Base; SI = Strict-IID; BI = BLOCK-5 IID; BS = BLOCK-5 with shared context; U = unrestricted with shared context; P = prior only; T = token only; CF = Full CF. P/T use strict closure. Shared context combines the correlated prior () and shared token; CF uses both under strict closure.
, : 2,000 episodes per configuration
| Task | B | SI | BI | BS | U | P | T | CF |
|---|---|---|---|---|---|---|---|---|
| S00 | 50 | 49 | 49 | 50 | 50 | 50 | 49 | 50 |
| S01 | 49 | 49 | 50 | 50 | 50 | 49 | 47 | 50 |
| S02 | 50 | 50 | 50 | 50 | 50 | 50 | 50 | 50 |
| S03 | 50 | 49 | 50 | 49 | 50 | 49 | 50 | 50 |
| S04 | 49 | 47 | 48 | 49 | 50 | 48 | 48 | 50 |
| S05 | 49 | 49 | 49 | 50 | 48 | 50 | 47 | 50 |
| S06 | 50 | 50 | 49 | 50 | 50 | 49 | 50 | 50 |
| S07 | 49 | 48 | 49 | 48 | 49 | 48 | 47 | 50 |
| S08 | 50 | 50 | 50 | 49 | 48 | 49 | 49 | 50 |
| S09 | 45 | 49 | 49 | 50 | 48 | 49 | 50 | 50 |
| O00 | 50 | 47 | 46 | 49 | 50 | 48 | 49 | 49 |
| O01 | 50 | 48 | 48 | 50 | 49 | 50 | 49 | 49 |
| O02 | 50 | 46 | 50 | 49 | 49 | 49 | 50 | 50 |
| O03 | 49 | 47 | 48 | 50 | 49 | 49 | 49 | 48 |
| O04 | 49 | 49 | 49 | 50 | 49 | 50 | 50 | 50 |
| O05 | 50 | 50 | 49 | 48 | 48 | 48 | 50 | 48 |
| O06 | 50 | 49 | 49 | 49 | 50 | 50 | 50 | 50 |
| O07 | 49 | 50 | 50 | 48 | 50 | 50 | 50 | 50 |
| O08 | 50 | 48 | 49 | 50 | 50 | 50 | 50 | 50 |
| O09 | 50 | 49 | 49 | 50 | 50 | 49 | 50 | 50 |
| G00 | 47 | 47 | 50 | 48 | 49 | 48 | 49 | 50 |
| G01 | 49 | 50 | 50 | 48 | 49 | 49 | 50 | 49 |
| G02 | 49 | 49 | 48 | 50 | 50 | 48 | 49 | 50 |
| G03 | 48 | 48 | 46 | 47 | 46 | 49 | 48 | 47 |
| G04 | 49 | 50 | 50 | 49 | 49 | 50 | 50 | 50 |
| G05 | 48 | 49 | 49 | 47 | 48 | 47 | 48 | 47 |
| G06 | 46 | 49 | 50 | 50 | 49 | 50 | 49 | 49 |
| G07 | 50 | 50 | 50 | 50 | 50 | 50 | 50 | 50 |
| G08 | 50 | 50 | 50 | 50 | 50 | 50 | 50 | 50 |
| G09 | 47 | 49 | 50 | 50 | 49 | 48 | 50 | 48 |
| L00 | 46 | 48 | 49 | 50 | 46 | 48 | 48 | 49 |
| L01 | 47 | 50 | 49 | 49 | 48 | 50 | 50 | 48 |
| L02 | 50 | 47 | 49 | 47 | 49 | 47 | 50 | 50 |
| L03 | 49 | 49 | 49 | 50 | 50 | 49 | 48 | 49 |
| L04 | 49 | 49 | 49 | 48 | 49 | 48 | 49 | 48 |
| L05 | 50 | 49 | 48 | 49 | 50 | 50 | 50 | 47 |
| L06 | 49 | 48 | 46 | 43 | 46 | 49 | 46 | 44 |
| L07 | 47 | 47 | 50 | 49 | 50 | 47 | 45 | 47 |
| L08 | 31 | 34 | 34 | 37 | 34 | 34 | 37 | 39 |
| L09 | 47 | 49 | 47 | 47 | 46 | 49 | 49 | 47 |
| Overall (%) | 96.80 | 96.70 | 97.15 | 97.30 | 97.20 | 97.20 | 97.45 | 97.60 |
: 2,000 episodes per configuration
| Task | B | SI | U | CF |
|---|---|---|---|---|
| S00 | 50 | 50 | 50 | 50 |
| S01 | 49 | 50 | 50 | 49 |
| S02 | 50 | 49 | 50 | 50 |
| S03 | 49 | 50 | 50 | 50 |
| S04 | 49 | 48 | 48 | 49 |
| S05 | 48 | 47 | 50 | 49 |
| S06 | 49 | 49 | 49 | 48 |
| S07 | 49 | 50 | 48 | 49 |
| S08 | 50 | 50 | 50 | 50 |
| S09 | 49 | 50 | 49 | 50 |
| O00 | 49 | 50 | 50 | 49 |
| O01 | 50 | 50 | 50 | 50 |
| O02 | 50 | 50 | 49 | 50 |
| O03 | 49 | 49 | 47 | 50 |
| O04 | 50 | 50 | 49 | 50 |
| O05 | 50 | 49 | 49 | 50 |
| O06 | 50 | 50 | 50 | 50 |
| O07 | 50 | 49 | 50 | 49 |
| O08 | 50 | 50 | 50 | 50 |
| O09 | 50 | 50 | 50 | 50 |
| G00 | 48 | 49 | 49 | 49 |
| G01 | 50 | 49 | 50 | 47 |
| G02 | 50 | 50 | 50 | 49 |
| G03 | 49 | 50 | 49 | 48 |
| G04 | 49 | 50 | 49 | 49 |
| G05 | 50 | 50 | 46 | 49 |
| G06 | 48 | 49 | 50 | 49 |
| G07 | 50 | 50 | 50 | 50 |
| G08 | 50 | 50 | 49 | 50 |
| G09 | 49 | 49 | 50 | 50 |
| L00 | 48 | 49 | 49 | 49 |
| L01 | 48 | 50 | 49 | 50 |
| L02 | 50 | 50 | 50 | 50 |
| L03 | 48 | 49 | 50 | 49 |
| L04 | 48 | 49 | 47 | 49 |
| L05 | 50 | 49 | 50 | 50 |
| L06 | 48 | 47 | 49 | 47 |
| L07 | 49 | 50 | 50 | 49 |
| L08 | 39 | 37 | 42 | 42 |
| L09 | 47 | 47 | 47 | 49 |
| Overall (%) | 97.90 | 98.15 | 98.15 | 98.30 |
Configuration and evaluation
The newly trained B/SI/U/P/T controls use initialization, seed 6113, 30,000 updates, global batch size 256, and 1,693 demonstrations. Optimization uses AdamW, a 10,000-update warm-up to , gradient clipping at norm 1, and EMA decay 0.999. CF and the two BLOCK-5 variants use existing trained checkpoints.
All configurations use ten Euler sampling steps. After ten settling actions, action budgets are 220/280/300/520 for S/O/G/L. The evaluation seed is 20261107. Tasks, initial states, environment-seed schedules, budgets, and official success criteria are shared across methods.
Normalization uses the saved training statistics. State dimensions are position (3), axis-angle orientation (3), and gripper joint positions (2). After inverse normalization and the output transformation, the first seven of 32 action coordinates provide translation deltas (3), rotation deltas (3), and gripper (1); the rest are padding. Commands are clipped to . Recorded actions are already deltas.
Reported 95% intervals use 10,000 paired task-bootstrap draws: sample ten tasks with replacement within each suite, apply the same task choices across methods, and retain each selected task's 50 episodes. Differences are in percentage points; bounds are the 2.5th and 97.5th percentiles.
Response to Reviewer MmNm — Part 1/2
Thank you for your careful review and constructive suggestions. Your questions helped us examine the boundary design and the two shared-context components more directly. We have added the requested matched evaluations and separate component ablations, with paired task-level uncertainty estimates. We present these results in Part 1, followed by reproducibility details and the interpretation of tail dependence in Part 2.
In a supplementary evaluation, we compared the boundary designs at H=10, K=5 on the same 40 LIBERO tasks and 50 official initial states per task, giving 2,000 episodes per configuration.
To compare the masks with the noise prior and shared-token setting held fixed, we also report two direct contrasts:
+0.45 pp [−0.30, 1.25].+0.30 pp [−0.35, 1.00].All 95% intervals are computed for the evaluated checkpoints using 10,000 paired task-bootstrap resamples, sampling tasks within each of the four LIBERO suites and retaining all 50 episodes of each selected task.
For a controller that always executes five actions per replan, BLOCK-5 retains bidirectional interaction within that block while isolating it from the discarded tail. When the controller may use different execution lengths, strict closure provides the invariance guarantee for every prefix. TRS directly shows this distinction:
We use 600 common observation contexts, holding observations, shared randomness, and prefix-local randomness fixed while resampling tail-local noise. This gives 2,400 intervention pairs per model at K=1 and 4,800 at K=5. BLOCK-5 has zero observed prefix change at its aligned K=5 boundary, while Strict-IID and Full CF retain zero observed change at both tested boundaries.
We added separately trained prior-only and token-only controls under the same strict dependency mask. Prior-only uses a correlated action-noise prior with ρ=0.5 and no shared token; token-only retains IID action noise and adds the shared token. Full CF combines both components. All four configurations use the same K=5 evaluation protocol.
Each component alone offsets the observed 0.10 pp decrease in Strict-IID's success rate relative to Base, and Full CF achieves the highest measured success rate. The Full-CF–prior-only difference is
+0.40 pp [−0.55, 1.35], and the Full-CF–token-only difference is+0.15 pp [−0.60, 0.90], using the same paired procedure.All four configurations retain zero observed TRS at K=1 and K=5. The strict dependency constraint provides prefix invariance, while the separate controls show how each coordination component affects measured task success under that constraint.