- Rating
- 3
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
MARS proposes training-free acceleration at three levels: Transformer components, diffusion denoising steps, and consecutive action chunks. The method has strong practical motivation and is evaluated on LIBERO, LIBERO-Pro, RoboCasa, several model backbones, multiple GPUs, and a real Franka robot. It achieves up to approximately 4× acceleration while largely preserving task success rates.
Critical strengths
MARS is a training-free acceleration framework for diffusion-based World Action Models. It combines calibrated attention skipping, action-dependent selection between a direct one-step output and residual-reuse sampling, and reuse of predicted visual latents after a shallow VAE consistency check. The method is mainly a coordinated combination of existing heuristic acceleration techniques, but the system-level integration and empirical results are valuable
Critical weaknesses
The main issues are that the redundancy calibration is primarily conducted on LIBERO, and it is unclear how the skip sets transfer to other benchmarks, backbones, and real-world environments. The visual-token reuse verifier lacks sufficient analysis of false positives, false negatives, reuse rates, and the action errors caused by incorrect reuse. Stability of DiT residuals does not necessarily imply stability of the final control actions, so trajectory-level and long-horizon failure analyses are still needed.
The visual verifier is not separated from reuse frequency and scheduling. Tables 4-5 do not include periodic or random reuse at a matched rate, or unconditional reuse under the same consecutive-reuse cap. The overall pipeline is empirically promising, but its results do not establish whether this verifier excludes harmful predictions better than simple scheduling.
Questions for authors
The authors should clarify calibration transfer, verifier overhead, and the relationship between feature stability and control safety.
3 comments
Re W2 & Q2: Verifier Reliability and Overhead
Benefits beyond refresh frequency. To test whether verification improves reuse decisions, we compared it with random reuse, periodic refresh, and unconditional reuse. At nearly identical skip rates (56.25% vs. 56.44%), our method improves success over random reuse from 44.87% to 45.53%, while achieving a 3.94× speedup. Periodic refresh and unconditional reuse skip more often but achieve lower success rates. The comparison suggests that choosing when to reuse cached computation matters. We have added these results and the corresponding analysis to Section 4.3 in the main paper.
| Method | Goal SR (%) / Speedup | Object SR (%) / Speedup | Spatial SR (%) / Speedup | Long SR (%) / Speedup | Success Rate ↑ | Speedup ↑ | Latency ↓ | Skip Rate |
|---|---|---|---|---|---|---|---|---|
| Cosmos Policy | 43.67 / 1.00× | 46.13 / 1.00× | 47.40 / 1.00× | 45.33 / 1.00× | 45.63% | 1.00× | 1.4044 s | 0.00% |
| Random Reuse | 42.87 / 4.09× | 45.80 / 4.11× | 46.80 / 4.09× | 44.00 / 4.20× | 44.87% | 4.14× | 0.3395 s | 56.44% |
| Periodic Refresh | 43.13 / 4.25× | 45.07 / 4.32× | 46.53 / 4.29× | 44.60 / 4.38× | 44.83% | 4.32× | 0.3249 s | 64.99% |
| Unconditional Reuse | 40.67 / 5.61× | 40.07 / 5.63× | 38.53 / 5.57× | 32.87 / 5.88× | 38.03% | 5.72× | 0.2457 s | 94.05% |
| Ours (Speed) | 44.20 / 3.91× | 45.67 / 3.86× | 47.33 / 3.96× | 44.93 / 3.99× | 45.53% | 3.94× | 0.3564 s | 56.25% |
Verification overhead. Verification requires shallow VAE encoding and change-map comparison. Its average cost is 0.89 ms per policy query, accounting for 0.25% of the total query time. Verification adds little overhead while enabling substantial speedups through visual token reuse.
| Verification Component | Time per Query (ms) | Share of Total Query Time (%) |
|---|---|---|
| Shallow VAE encoding | 0.59 | 0.17 |
| Change-map comparison | 0.30 | 0.08 |
| Total verification | 0.89 | 0.25 |
Re W3 & Q3: Feature Stability and Control Safety
We thank the reviewer for emphasizing the connection between feature stability and control performance.
Feature Stability and Control Safety. We directly compared residual reuse with full multi-step inference on 40 LIBERO-PRO tasks. We collected queries from complete rollouts of both policies. For each query, both methods used the same observation, initial noise, and sampling schedule. For each query, we compute the L2 differences between the accelerated and original actions. The table shows that residual reuse closely preserves the original action outputs, with small translational and rotational deviations and 99.14% gripper agreement.
| Method | Translation L2 ↓ | Rotation L2 ↓ | Gripper agreement (%) ↑ |
|---|---|---|---|
| Original Cosmos Policy | 0.0000 | 0.0000 | 100.00 |
| Residual Reuse | 0.0296 | 0.0051 | 99.14 |
Trajectory-level evidence. As analyzed across 12,000 paired episodes in Appendix G, our method yields a gain in overall success rate, frequently recovering trajectories involving direct object transport (Figure 5). Conversely, it is slightly more sensitive to precision-critical interactions (Figure 6).
Long-horizon performance. Table 1 also shows that Ours (Acc.) maintains success rates close to the base model on LIBERO-Long and LIBERO-PRO-Long, with 3.19× and 3.11× speedups, respectively.
| Benchmark | Cosmos Policy SR (%) | Ours (Acc.) SR (%) | Speedup |
|---|---|---|---|
| LIBERO-Long | 96.67 | 96.60 | 3.19× |
| LIBERO-PRO-Long | 45.33 | 44.60 | 3.11× |
We hope these results address the reviewer's concerns, and we would be grateful if the reviewer would consider raising the score.
Dear Reviewer,
As the discussion period is nearing its end, we would greatly appreciate your feedback on whether our rebuttal has addressed your concerns. We hope our clarifications can be considered in your final assessment.
Thank you for your time and consideration.
Re W1 & Q1: Calibration Transfer
We thank you for highlighting the need to clarify the transferability of calibration.
Transfer across benchmarks and backbones. To test configuration generality, we conducted additional experiments using shared configurations. These shared-configuration experiments were conducted on RTX 4090 GPUs, with results averaged over three seeds.
Across benchmarks, we use one fixed configuration on LIBERO, LIBERO-PRO, and RoboCasa without benchmark-specific tuning.
Across backbones, we also evaluate a shared configuration on Fast-WAM, LingBot-VA, and π₀.₅ without backbone-specific tuning.
These results show that shared configurations can reduce latency while maintaining comparable success rates across the evaluated benchmarks and backbones. We have added a new section, Appendix J, to describe these experiments and present the results in Tables 12 and 13.
The real-robot task. For real-world deployment, we directly reuse the exact hyperparameter configuration from simulation without any modification, demonstrating seamless sim-to-real transferability with zero additional tuning effort.