Vision-language-action (VLA) models increasingly pair a pre-trained vision-language model (VLM) with a flow-matching expert that conditions its action predictions on the VLM's key-value representations. Those representations must carry information useful for control, yet the VLM was never trained for the action expert's flow-matching objective. Two signals can address this: gradients backpropagated from the expert's loss (coupling), or VLM supervision to predict discrete tokens encoding the same target actions. Prominent training recipes forgo the first, blocking expert-to-VLM gradients (insulation) to protect the VLM's existing capabilities, while also supervising it on vision-language data and action tokens. Because vision-language co-training also aims to preserve those capabilities, the additional protection provided by insulation is unclear. At the same time, insulation removes a direct signal about which features continuous action prediction requires. On the other hand, action-token supervision offers an alternative way to learn action-relevant representations, which may explain why insulation succeeds in practice. We disentangle these factors in autonomous driving and manipulation by independently varying action-token supervision and whether the expert's loss updates the VLM. In open-loop driving, insulation is competitive with coupling when action tokens are supervised; without them, coupling reduces trajectory-prediction error by roughly 25%. When both components are trained jointly from the outset, coupling increases closed-loop driving route progress by approximately 17% with action-token supervision and 40% without it, and similarly improves robot manipulation performance under both settings. These findings show that the value of gradient insulation depends on the VLM's action supervision and the evaluation setting, challenging insulation as a default VLA training recipe.
Figures & tables
Figure 1: Action-token supervision shapes the effect of expert-to-VLM gradient flow. Left: the VLM is supervised to predict discrete action tokens while a flow-matching expert decodes continuous actions from the VLM’s K/V cache. We independently vary whether action tokens are supervised and whether the expert’s gradient ( ∇Lflow ) updates the VLM. Right: open-loop driving evaluation error during training. With action-token supervision, insulated and coupled training differ little; without it, coupling is substantially better. We test this interaction in closed-loop autonomous driving and robot manipulation.
Policy
VQA Predicted
Discrete Actions Predicted
Autonomous Driving
Alpamayo-R1 [ 32 ]
✓
✓
Qwen-Drive-1.0 [ 43 ]
✓
–
SpanVLA [ 46 ]
–
✓
UniDriveVLA [ 22 ]
✓
–
Manipulation
Table 1: Driving and manipulation VLAs that block action-expert gradients from the VLM through freezing or knowledge insulation [ 6 ] . Checkmarks indicate VQA or discrete action-token supervision.
Figure 2: Action-token supervision, gradient routing, and training schedules. Left : we cross future-action-token supervision with coupled or insulated expert-to-VLM gradients. Right : bars show when the VLM (blue) and action expert (black) are trained. Concurrent training starts both modules together; VLM-first and expert-first training start one module after 100k steps of training the other module. All schedules run for 300k total steps in autonomous driving experiments.
Figure 3: Open-loop driving accuracy across training histories. First-sample mean ADE on the Public Golden (1,434 examples) and OOD (273 examples) evaluation splits. Bars show the mean across scenes; error bars show 95% clip-cluster bootstrap confidence intervals. Lower is better.
Public Golden ( N=1,434 )
OOD ( N=273 )
Schedule
Routing
Action CE
Top-1 [95% CI]
Top- k
Avg- k
Top-1 [95% CI]
Top- k
Avg- k
Concurrent
Insulated
On
0.647 [0.621,0.673]
0.808
0.630
0.414 [0.355,0.474]
0.659
0.466
Concurrent
Coupled
On
0.649 [0.623,0.674]
0.823
0.629
0.425 [0.364,0.487]
0.689
0.471
Concurrent
Insulated
Off
0.566 [0.539,0.593]
0.767
0.573
0.363 [0.305,0.423]
0.652
0.435
Concurrent
Coupled
Off
0.635 [0.610,0.661]
0.811
0.622
0.348 [0.292,0.406]
0.685
0.450
VLM-First
Insulated
On
0.639 [0.614,0.665]
0.807
0.626
0.392 [0.335,0.452]
0.652
0.462
Table 2: Generated reasoning-trace quality. First-sample Lingo-Judge pass rates for each model. In each example, the frozen Lingo-Judge classifier compares the reasoning-trace with a reference rationale [ 27 ] ; an example passes when the judge score exceeds 0.5 . Brackets give 95% clip-cluster bootstrap confidence intervals. Top- k is the fraction for which any of six sampled rationales passes, and Avg- k is the mean judge score across the six samples. Results are evaluated on Public Golden and OOD. Within each training schedule, the best metric is bolded. Higher is better.
Insulated
Coupled
History
Tokens on
Tokens off
Tokens on
Tokens off
Concurrent
0.600
0.425
0.703
0.601
VLM-First
0.648
0.555
0.646
0.587
Expert-First
0.597
0.538
0.625
0.604
Table 3: Closed-loop route progress across training histories. Mean AlpaSim route progress on the same 827 scenes, with five stochastic rollouts per scene averaged within each scene. The best value in each row is bolded and the second-best is underlined. Higher is better.
Condition
π0.5
MolmoAct2
Insulated / FAST on
19.5%
22.0%
Coupled / FAST on
30.7%
24.3%
Insulated / FAST off
10.3%
14.5%
Coupled / FAST off
32.9%
36.5%
Insulated / FAST on
40.0%
32.9%
Table 4: Closed-loop manipulation progress. Equal-weight mean normalized task progress across three representative PolaRiS tasks. Bold indicates the best result among our four conditions within each model family. The final row provides additional references: the published π0.5 result is from CoVeR [ 20 ] and uses its evaluation protocol; the MolmoAct2 result uses its released checkpoint with our own fine-tuning and evaluation scheme. Higher is better.
Appendix figures & tables9 assets
Supplementary material from the paper’s appendix.
Appendix
Setting
Value
Optimizer
Fused AdamW
(β1,β2)
(0.9,0.95)
Adam ϵ
10−8
Weight decay
0.1
Vision LR
10−5
VLM text/head LR
10−4
Appendix
Table 5: Driving optimization settings. Shared across conditions within each training history.
History
Schedule
VLM updates
Expert updates
Concurrent
300k joint steps
300k
≈145.5 k
Expert-First
100k expert-only + 200k joint
200k
≈197 k
VLM-First
100k VLM-only + 200k joint
300k
≈97 k
Appendix
Table 6: Expected component updates. Endpoints have 300k total steps
Component
Backbone
Layers
Width
MLP
Heads (Q/KV)
π0.5
VLM Backbone
Gemma 2B
18
2,048
16,384
8/1
Action expert
Gemma 300M expert (random)
18
1,024
4,096
8/1
MolmoAct2
VLM Backbone
Molmo2-ER
36
2,560
19,456
32/8
Action expert
MolmoAct2 expert (random)
36
768
3,072
8/8
Appendix
Table 7: Manipulation model dimensions. Attention heads are query/key-value counts.
Setting
Value
Optimizer
AdamW
(β1,β2)
(0.9,0.95)
Adam ϵ
10−8
Weight decay
10−10
Peak/terminal LR
5×10−5
Warm-up
1,000 steps
Appendix
Table 8: π0.5 manipulation optimization settings. Shared across conditions through 400k steps.
Component
Peak LR
Language model
1×10−5
Vision encoder
5×10−6
Connector
5×10−6
Action expert
5×10−5
Appendix
Table 9: MolmoAct2 learning rates. Shared across conditions and used for both training and adaptation.
Figure 4: Closed-loop manipulation examples. Rollouts of the four concurrently trained policies in PanClean (left), MoveLatteCup (center), and BlockStackKitchen (right). MoveLatteCup is an additional qualitative example and is excluded from the three-task averages. Check marks indicate task success and crosses indicate failure. Among these examples, only the coupled policies succeed.
#
Criterion
Condition
Requires
PanClean
1
Reach sponge
End effector within 0.2m of sponge
—
2
Lift sponge
Sponge height >0.09m
1
3
Sponge on pan
At least 80% overlap with pan; gripper open
2
FoodBussing
1
Reach ice cream
End effector within 0.2m
—
Appendix
Table 10: PolaRiS rubric criteria. Overlap is measured between horizontal bounding boxes. Placement requires the gripper to be open.
Figure 5: Open-loop driving across training histories. Selected-trajectory minADE on the Public Golden dataset and OOD evaluation dataset. Bars show condition means, and error bars show 95% clip-cluster bootstrap confidence intervals for those means. Lower is better.
Figure 6: Concurrent-training AlpaSim case study. Four conditions in the same traffic scene. The insulated rollout without action-token supervision has an at-fault collision. The displayed rollouts for the other three conditions succeed. This example illustrates a failure mode behind the aggregate results in Table 11 .
School of Information, Renmin University of China, Beijing, China · Key Laboratory of Data Engineering and Knowledge Engineering, Beijing, China · Engineering Research Center of Database and Business Intelligence, Beijing, China