- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper presents Anchor-Align, a VLA finetuning method that augments behavior cloning with two objectives. Vision-Language Anchoring matches intermediate representations to those of a frozen pretrained VLM, while Language-Action Alignment supervises language predictions using motion-direction labels derived from robot demonstrations. Together, these objectives aim to preserve pretrained knowledge and connect it to action learning. Experiments on LIBERO-PRO, LIBERO-Plus, CALVIN, and real-world single-arm and bimanual tasks report improvements in generalization and manipulation performance across regression and flow-matching action heads, without requiring external data or additional manual annotations.
Critical strengths
- Simple and practical method. Both auxiliary objectives use existing robot demonstrations, requiring no additional annotations, external training data, or language generation at inference.
- Real-world coverage. Real-world experiments include single-arm and bimanual robots, as well as regression and flow-matching action heads, providing evidence of applicability across different configurations.
Critical weaknesses
- Limited novelty. Preserving VLM representations and using language supervision for action learning have both been explored widely in prior work. The main changes here are layer-wise feature distillation and supervision with coarse motion-direction labels, which appear to be relatively small extensions of existing ideas.
- Comparison with stronger pretrained VLAs. The paper would benefit from comparisons with models such as π0 and π0.5 on both real-world tasks and simulation OOD benchmarks. While Appendix Table 5 includes π0.5-KI on standard LIBERO, evaluating these models under the same downstream demonstrations and OOD conditions would better establish how the proposed method compares with existing pretrained VLAs in generalization.
- Overstated claims in the introduction. Several claims in the introduction appear broader than the evidence supports, including that prior methods are “never supervised on the same robot observation” (L94–102) and that “losses on such auxiliary data do not transfer the VLM’s semantic understanding to the finetuned VLA” (L89–90). Approaches such as π0.5 and Knowledge Insulation already combine token prediction and continuous-action learning on robot observations, including action-related language supervision. Moreover, π0.5’s web-data ablations demonstrate benefits for language following on unseen object categories, raising the possibility that the negative results reflect issues with the experimental setup or baseline implementation, rather than a general failure of co-training to transfer semantic knowledge. The paper would benefit from narrowing these claims, distinguishing observed failures from their proposed explanations, and clarifying its specific contribution beyond existing joint supervision and knowledge-preservation methods.
Questions for authors
- Is the method actually leveraging pretrained language semantics, or simply learning an additional action-direction classifier?
- See Weakness.
5 comments
Author Response (2/3)
Comparison with stronger pretrained VLAs. The paper would benefit from comparisons with models such as and on both real-world tasks and simulation OOD benchmarks.
A key distinction in this comparison is that Anchor-Align (based on StarVLA / VLA-Adapter backbone) requires no action pretraining. It directly adapts a pretrained VLM using only downstream demonstrations, whereas & benefit from large-scale robot pretraining (L118–121). The comparison allows us to assess how far this finetuning recipe can go compared to a strong robot-pretrained model under the same downstream training data budget. In real-world settings, we only compare with because: 1) simulation results show that Anchor-Align is clearly better than , and 2) real-world experimentation is very time-intensive and labor-intensive.
Real-world comparison. Anchor-Align's main strength comes from preserving and leveraging the VLM's pretrained representations, which are particularly useful for real-world observations due to large-scale vision-language pretraining. Our real-world results demonstrate this advantage: Anchor-Align outperforms across 6 out of 8 tasks, particularly on challenging tasks involving Motion Understanding, Stability, and Target Generalization (OOD).
Extension of Figure 6(d) in the main paper
| Method | Reactive Grasp | │ | Tomato Plate Transfer | │ | Tomato Plate OOD Transfer | ||
|---|---|---|---|---|---|---|---|
| Success (%) | │ | Placed (%) | Retained (%) | │ | Placed (%) | Retained (%) | |
| Standard BC | 50 | │ | 5 | 6.3 | │ | 0 | 0 |
| [new] | 40 | │ | 70 | 37.5 | │ | 40 | 18.75 |
| Anchor-Align | 70 | │ | 100 | 60 | │ | 70 | 47.5 |
Extension of Figures 7, 9(b), and 10 in the main paper (StarVLA results)
| Method | Pink Mug Placement | │ | Cable Insertion | │ | Green Mug Placement | │ | Broccoli | │ | Clutter | |
|---|---|---|---|---|---|---|---|---|---|---|
| Pickup (%) | Placed (%) | │ | Success (%) | │ | Success (%) | │ | Success (%) | │ | Success (%) | |
| Standard BC | 0 | 0 | │ | 10 | │ | 30 | │ | 35 | │ | 45 |
| [new] | 90 | 25 | │ | 25 | │ | 50 | │ | 60 | │ | 75 |
| Anchor-Align | 100 | 40 | │ | 50 | │ | 60 | │ | 50 | │ | 70 |
Simulation comparison. We extend our simulation evaluation (Table 1) to include and baselines. Anchor-Align substantially outperforms and achieves competitive performance with in 8 out of 10 OOD categories. Anchor-Align underperforms on Position Swap and Initial Robot State, two categories where strong action priors are particularly beneficial. This makes 's advantage in these settings intuitive, given its extensive robot pretraining.
Importantly, Anchor-Align is a finetuning recipe that can also be applied directly to . Applying Anchor-Align to improves performance from 75.1% to 78.3% on LIBERO-PRO and from 93.8% to 94.4% on LIBERO-Plus. Notably, performance improves on Position Swap (32.8% → 41.2%) and Initial Robot State (79.4% → 82.9%), two challenging categories where large-scale action pretraining provides an advantage. On LIBERO-Plus, already achieves at least 94% accuracy in five of seven categories, leaving limited room for further improvements. These results demonstrate that Anchor-Align further improves , establishing the generalizability of our finetuning recipe across different VLA backbones and its complementarity to large-scale robot pretraining.
Extension of Table 1 in the main paper
| Method | Lang. Reph. | Object Swap | Pos. Swap | PRO Mean | Lang. Instr. | Bg. Text. | Robot Init | Cam. View | Obj. Layout | Light Cond. | Sensor Noise | Plus Mean |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MolmoAct | 77.8 | 82.4 | 0.0 | 53.4 | 79.5 | 84.1 | 47.4 | 10.1 | 76.5 | 77.4 | 53.4 | 59.9 |
| OpenVLA-OFT | 74.4 | 95.2 | 0.0 | 56.5 | 81.5 | 95.7 | 40.3 | 94.7 | 88.6 | 95.5 | 28.2 | 74.1 |
| [new] | 92.0 | 95.4 | 0.0 | 62.5 | 83.8 | 96.1 | 37.1 | 95.5 | 91.9 | 96.2 | 87.7 | 83.6 |
| VLA-Adapter | 91.1 | 89.6 | 2.3 | 61.0 | 85.1 | 90.7 | 52.6 | 92.6 | 93.2 | 93.2 | 89.5 | 85.1 |
| Anchor-Align VLA (ours) | 97.0 | 96.2 | 22.6 | 71.9 | 87.2 | 99.6 | 59.1 | 96.3 | 97.4 | 99.0 | 96.9 | 90.3 |
| [new] | 95.8 | 96.8 | 32.8 | 75.1 | 96.2 | 100.0 | 79.4 | 92.0 | 97.4 | 99.3 | 94.0 | 93.8 |
| Anchor-Align- (ours) [new] | 96.4 | 97.4 | 41.2 | 78.3 | 96.4 | 99.2 | 82.9 | 94.4 | 97.4 | 100.0 | 92.0 | 94.4 |
Author Response (3/3)
W3
Overstated claims in the introduction.
“never supervised on the same robot observation” (L94–102)
Thank you for pointing it out. We fixed it in the revision, clarifying our contribution.
“losses on such auxiliary data do not transfer the VLM’s semantic understanding to the finetuned VLA” (L89–90)” … Moreover, ’s web-data ablations demonstrate benefits for language following on unseen object categories, raising the possibility that the negative results reflect issues with the experimental setup or baseline implementation, rather than a general failure of co-training to transfer semantic knowledge.
We think there is some confusion here; please allow us to clarify. The reviewer is correct that ’s web-data ablations demonstrate improved language following on unseen object categories through its large-scale training recipe. However, our study addresses a different setting, downstream VLA finetuning, where our experiments show that auxiliary co-training interferes with learning task-relevant action representations. Similar limitations have been independently reported by ECoT [9], ChatVLA [10], and RAFT [11]. These findings do not contradict ; they demonstrate that the effectiveness of auxiliary supervision depends on the training regime and how it is incorporated.
The following is references from our submission that discuss this hindrance and empirically report performance in such cases. In Appendix A, L815–820, we discuss why this might be the case:
We hypothesize that general-purpose web supervision can act as an auxiliary objective with limited relevance to these task details, potentially competing with task-specific adaptation. This may help explain why our Co-training + KI baseline underperforms standard BC on mean LIBERO-PRO and LIBERO-Plus success.
Our representation analysis (Appendix D.4) further shows that preserving pretrained feature structure does not necessarily yield action-informative representations: the co-trained backbone retains high representational similarity but exhibits lower action decodability than the action-trained alternatives.
Q1
Is the method actually leveraging pretrained language semantics, or simply learning an additional action-direction classifier?
During the rebuttal phase, we conducted a new experiment: we remove the frozen pretrained language head and replace it with a classifier that directly predicts the six motion directions. We use the same direction labels derived from demonstrated action chunks, so the classifier’s supervision remains aligned with action generation. We train this aligned-action classifier jointly with BC and anchoring, and its loss updates both the classifier and the backbone’s trainable LoRA parameters.
| Method | PRO | Plus |
|---|---|---|
| VLA-Adapter | 61.0 | 85.1 |
| Aligned-Action Classifier [new] | 65.5 | 87.8 |
| Anchor-Align VLA | 71.9 | 90.3 |
Anchor-Align outperforms this classifier, supporting the benefit of pretrained language semantics beyond action-aligned direction supervision.
When we examine the VLM backbone’s understanding of motion directions (Appendix, Table 8), we find that Anchor-Align strengthens the motion semantics encoded in the backbone, with gains that transfer beyond the training distribution. We train linear probes on LIBERO-Spatial and evaluate them on both LIBERO-Spatial and the out-of-distribution LIBERO-Object suite. The probes operate on fixed backbone representations and bypass both the learned projection and the frozen language head, directly assessing directional information within the backbone.
Table 8 copied from appendix
| Method | Spatial (in-dist.) | Object (OOD) |
|---|---|---|
| Frozen VLA-Adapter backbone | 43.8 | 22.8 |
| VLA-Adapter | 86.0 | 21.4 |
| Anchor-Align VLA | 94.4 | 30.3 |
Author Response (1/3)
W1
Preserving pretrained VLM representations while adapting them for continuous control during VLA finetuning is largely understudied. To our knowledge, Anchor-Align is the first VLA finetuning objective to jointly preserve pretrained VLM representations and align them with action, without additional data.
The adaptation of the VLM backbone during finetuning creates a preservation–adaptation tension: finetuning on action causes catastrophic forgetting of VLM priors, whereas strongly preserving them (e.g., by freezing the VLM backbone) does not yield representations useful for control. Prior work mainly focuses on large-scale pretraining [1][2][3], showing how language supervision and preserving VLM representations through co-training can help. We show that these methods do not naturally transfer to finetuning. VLAs remain fragile under distribution shifts [6][7], and prominent approaches use additional pretraining data to improve generalization [2][8]. Here, we show that this is an effect of standard BC finetuning itself (L44–53), and co-training objectives can introduce task interference (further supported by [4][5]).
Anchor-Align addresses both within the finetuning objective, using supervision already available in robot demonstrations. Anchoring constrains representational drift while allowing both the continuous-action BC loss and the language–action alignment loss to adapt the VLM backbone (Figure 2 caption, L81–85). The implementation is simple; the gains are significant and consistent across 4 simulation benchmarks: LIBERO-PRO, LIBERO-PLUS, LIBERO, and CALVIN, and 8 real-world manipulation tasks spanning different embodiments (YAMs and XARM) and different backbones. The additional gains when finetuning π0.5 demonstrate that Anchor-Align complements large-scale pretraining (author response 2/3).