- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes StairVLA, a hierarchical VLA framework that splits flow-matching action generation into two stages. A high-level policy generates partially denoised long-horizon action trajectories, which are cached and reused across multiple control cycles. A lightweight refiner then uses updated observations to refine local action chunks. Experiments on LIBERO, LIBERO-Plus, and real-world robot tasks show reduced inference latency while maintaining competitive success rates.
Critical strengths
-
Using partially denoised actions as the interface between high-level and low-level policies is an interesting idea. It avoids introducing explicit subgoals or other handcrafted intermediate representations.
-
The method achieves considerable inference speedups while maintaining or slightly improving success rates on both GR00T-style and pi-style backbones.
-
The evaluation covers multiple benchmarks, real-world experiments, and several useful ablations. The analysis of denoising stages and trajectory reuse also helps motivate the design.
Critical weaknesses
-
The role of partially denoised actions is not sufficiently isolated. The refiner also receives compressed high-level VLM features, which already contain task-related information. It is therefore unclear how much task intent is actually transferred through the intermediate actions. The current ablations do not resolve this.
-
The advantage over simpler coarse-to-fine approaches remains unclear. A high-level policy could generate a fully denoised trajectory, followed by a lightweight observation-conditioned refiner. The paper does not compare against this alternative, although the inference-time denoising level is already close to completion. This makes it difficult to assess the specific benefit of splitting the denoising process.
-
The evidence for stage-dependent modality reliance is not entirely convincing. The attention trends differ between GR00T and pi-style models, and the attention analysis uses only 16 samples. The results suggest that conditioning changes across denoising stages, but do not firmly establish the proposed division between early task understanding and late visual refinement.
-
The performance and efficiency comparisons are not fully controlled. The baselines use 8-action chunks while StairVLA uses 5-action chunks, making the reported per-chunk latency less directly comparable. StairVLA also introduces an additional refiner and training stage, so some performance gains may come from increased capacity or training rather than the proposed mechanism. The real-world results are encouraging, but the limited number of trials makes the improvements less conclusive.
Questions for authors
-
Can the authors provide an ablation removing the cached high-level VLM features to clarify how much task information is carried by the partially denoised actions?
-
How does StairVLA compare with a baseline that uses fully denoised long-horizon trajectories followed by the same observation-conditioned refiner? This seems important for establishing the benefit of the proposed interface.
-
Can the authors provide stronger evidence for the claimed stage-dependent modality reliance, particularly given the different attention patterns observed across backbones?
-
Can the authors report latency normalized by executed actions and account for high-level policy refresh overhead? It would also be useful to clarify how much of the improvement remains under matched model capacity and training budgets.
3 comments
Response to Reviewer JpA3 (2)
Q4: Fair latency comparison, model capacity, and training budget.
We use on LIBERO primarily to match the action chunk size of Libra-VLA for a fairer comparison.
| Method | Chunk Size | Success (%) | ms/chunk | ms/action |
|---|---|---|---|---|
| StarVLA-GR00T | 8 | 96.50 | 115.0 | 14.38 |
| Libra-VLA | 5 | 97.20 | 122.0 | 24.40 |
| StairVLA (Ours) | 5 | 97.80 | 44.2 | 8.84 |
StairVLA reduces inference latency per executed action by 38.5% compared with StarVLA-GR00T.
In our real-world Fruit25 experiments, we use , matching the StarVLA baseline. The results were originally reported in Figure 10; we additionally provide per-action latency below.
| Method | Chunk Size | Success (%) | ms/chunk | ms/action |
|---|---|---|---|---|
| StarVLA-GR00T | 8 | 72.8 | 114.9 | 14.36 |
| StairVLA (Ours) | 8 | 78.4 | 65.7 | 8.21 |
On Fruit25, StairVLA reduces per-action inference latency by 42.8% compared with StarVLA-GR00T.
Model capacity and training budget. Amortized over executed actions, StairVLA incurs lower average inference computation cost. To examine whether the performance gains arise merely from additional model capacity or training, we conduct a Deep GR00T ablation, doubling the action head from 16 to 32 layers. We follow the same two-stage training schedule as StairVLA, with 30k steps per stage; in the second stage, the VLM is frozen while the expanded action head is trained.
| Method | Action Head | Training Steps | Success (%) | (pp) |
|---|---|---|---|---|
| StarVLA-GR00T | 16 layers | 30k | 96.50 | -- |
| Deep GR00T | 32 layers | 30k + 30k | 95.70 | -0.80 |
| StairVLA (Ours) | 16 layers + refiner | 30k + 30k | 97.80 | +1.30 |
These results suggest that simply increasing model capacity and training budget does not reproduce the performance gains of StairVLA, supporting the effectiveness of our hierarchical action generation design.
The corresponding experiments and analyses have been incorporated into Table 3 and the latency analysis in the revised manuscript and appendix.
Gentle Follow-up on Our Rebuttal
Just a gentle follow-up on our rebuttal and additional experiments. We would really appreciate any feedback or further questions you may have.
Thank you again for your time and consideration!
Response to Reviewer JpA3 (1)
We greatly appreciate the reviewer's positive feedback on the "interesting idea" of using partially denoised actions as the high-to-low-level interface, "considerable inference speedups", and the extensive evaluation across multiple benchmarks and real-world experiments. We address the remaining concerns below.
We conduct two additional ablation experiments, with the results summarized below.
The ablation results indicate that partially denoised actions can convey useful task information even without cached VLM features. Nevertheless, we have revised the statement in Section 1 to more carefully describe the role of partially denoised actions in transferring task information, alongside cached VLM features.
Why are VLM features alone insufficient? The VLM-features-only setting resembles conventional hierarchical VLA designs, which often require additional training mechanisms or intermediate supervision to effectively guide low-level policies.
Why is fully denoised refinement less effective? The strong fitting capability of the high-level VLA may limit effective residual learning after full denoising: although increasing GR00T's action horizon from 8 to 32 reduces closed-loop success from 96.5% to 92.4%, the per-action RMSE evaluated on training samples decreases from 0.0431 (H=8) to 0.0237 (H=32).
The corresponding experimental results and analysis have also been included in Table 3 of the revised manuscript.
We evaluate instruction and image masking at different denoising stages on GR00T over 400 LIBERO episodes. Here, s denotes the masking onset, with the corresponding conditioning masked from s to 1. The table reports success-rate changes (percentage points) relative to normal inference.
Late-stage denoising remains critical. Masking images during only the final 10% of denoising reduces success by 91.00 points, causing almost complete failure.
Visual conditioning dominates late-stage sensitivity. At 0.97, instruction masking reduces success by only 1.00 point, while image masking still causes a 7.25-point drop, consistent with our empirically selected α=0.97.
See revised Section 4.3 and Appendix B.1.4 for detailed results across both backbones. Although the patterns differ across backbones, the π-style attention analyses (repeated with 1,024 samples, with consistent findings) and GR00T interventions jointly motivate reducing repeated language processing while retaining observation-conditioned refinement. Our main results further validate this design's performance--efficiency benefits.