- Rating
- 3
- Confidence
- 2
- Informativeness
- 3
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
AIDA attacks the initial visual observation of a latent world model. It makes predictions under different actions more similar while pushing them away from clean predictions. A third loss suppresses the state-transition Jacobian’s spectral norm to limit how action-induced differences grow during rollout. The paper tests the attack on LeWM, DINO-WM, and PLDM using PushT and OGBench Cube.
Critical strengths
-
The attack targets a capability that planners depend on: distinguishing the consequences of different actions. Making one prediction inaccurate may leave other useful plans available; reducing differences across actions attacks the comparison itself. The sensitivity analysis gives a reason for including the propagation loss.
-
The attack achieves the lowest task success rate in all six tested model–environment combinations. On Cube, LeWM success falls from 79.0% to 24.2%, and DINO-WM from 80.0% to 29.6%. The loss ablations and action-sensitivity measurements help explain the results, while the MPPI evaluation checks that the effect is not limited to CEM.
Critical weaknesses
-
The information available to the attacker is unclear. Section 4.1 uses recorded future actions to construct action ensembles and subsequent clean observations for teacher-forced prediction. Access to model parameters and gradients does not by itself provide this information. The paper should state whether these trajectories are assumed to be available and what attack setting this represents. The evaluation also needs clarification: Cube cases are filtered for clean LeWM success, yet the reported clean success rate is 79%, and the separation between hyperparameter tuning and final evaluation is not clear.
-
The results do not establish that action discriminability remains impaired after the controller receives fresh observations. Equation (8) and Figure 6 concern predictions within a latent rollout. Figure 5 shows that task performance remains affected under different replanning settings, but this could also reflect damage caused by the initial incorrect actions. Both effects matter, but they are different explanations. The paper should distinguish them and either add evidence after clean observations are reintroduced or narrow the persistence claim.
Questions for authors
-
Does attack generation require the recorded future actions and the three subsequent clean observations? If so, how are these made available to the attacker? Can the attack also be constructed using only current and past observations?
-
Why is clean LeWM success on Cube 79% after filtering for clean success? Please clarify whether filtering and evaluation use different seeds or settings, whether clean and attacked runs are paired, and whether tuning and evaluation use separate data.
-
After a fresh clean observation enters the controller, does the model still produce less distinguishable predictions across actions, or is the remaining performance loss mainly due to the earlier actions?
-
For the propagation loss, which history variables are differentiated, and at which prediction steps is the Jacobian evaluated?
5 comments
Response to reviewer fC8v regarding to Weakness 1, Questions 2 (Part 2/3)
Evaluation
"W1: ... The evaluation also needs clarification: Cube cases are filtered for clean LeWM success, yet the reported clean success rate is 79%, and the separation between hyperparameter tuning and final evaluation is not clear." and "Q2: Why is clean LeWM success on Cube 79% after filtering for clean success? Please clarify whether filtering and evaluation use different seeds or settings, whether clean and attacked runs are paired, and whether tuning and evaluation use separate data."
Thank you for pointing out these ambiguities. Clean-success filtering was performed on a larger pool of episodes, while final evaluation reran only the selected successful episodes. Despite using the same seed, the different episode sets led to different random number generator states and CEM samples, resulting in a 79% clean SR. Importantly, clean and attacked evaluations are paired, using identical random seeds and random sampling sequences for each episode. To further ensure that our results are not specific to a particular evaluation seed, we further evaluate AIDA, TRAP and UADA/TMA using five different random seeds. Results are shown in the table below:
| Method | SR (%) ↓ | ASR (%) ↑ | FGD (cm) ↑ |
|---|---|---|---|
| TRAP | 74.20 ± 2.02 | 21.20 ± 1.83 | 5.891 ± 0.261 |
| UADA/TMA | 39.90 ± 1.71 | 57.47 ± 1.89 | 9.90 ± 0.26 |
| AIDA (Ours) | 26.00 ± 4.17 | 71.57 ± 5.29 | 14.35 ± 0.68 |
Across 5 seeds, ours consistently outperforms UADA/TMA. These results suggest that AIDA's effectiveness is not tied to a particular evaluation seed.
We acknowledge that the sensitivity analyses of loss weights and action-sampling noise scale were conducted on the original evaluation set. To verify that the selected hyperparameters generalize beyond this set, we additionally evaluate AIDA and other baselines on 200 new test cases for each dataset, using the same hyperparameters without further tuning.
| World Model | Method | OGBench-Cube | PushT | |||||
|---|---|---|---|---|---|---|---|---|
| SR ↓ | ASR ↑ | FGD ↑ | SR ↓ | ASR ↑ | FPE ↑ | FAE ↑ | ||
| LeWM | Clean | 83.00 | – | 5.00 | 86.00 | – | 28.43 | 6.51 |
| TRAP | 76.50 | 16.87 | 5.48 | 80.50 | 13.37 | 26.44 | 6.82 | |
| UADA/TMA | 39.50 | 58.43 | 9.56 | 10.50 | 87.79 | 255.51 | 23.13 | |
| Ours | 25.50 | 70.48 | 14.07 | 9.00 | 89.53 | 245.46 | 22.76 | |
| DINO-WM | Clean | 79.00 | – | 4.53 | 38.00 | – | 38.28 | 13.75 |
| TRAP | 80.50 | 16.46 | 4.66 | 23.50 | 52.63 | 51.42 | 20.25 | |
| UADA/TMA | 51.00 | 46.20 | 7.75 | 17.00 | 68.42 | 57.83 | 25.47 | |
| Ours | 33.50 | 63.29 | 10.23 | 12.50 | 75.00 | 83.31 | 29.71 | |
| PLDM | Clean | 46.50 | – | 8.69 | 74.00 | – | 93.00 | 10.58 |
| TRAP | 42.00 | 26.88 | 8.90 | 72.00 | 9.46 | 82.54 | 11.43 | |
| UADA/TMA | 35.00 | 52.69 | 10.31 | 13.00 | 82.43 | 266.44 | 27.03 | |
| Ours | 31.00 | 59.14 | 11.98 | 8.00 | 90.54 | 325.48 | 29.09 |
Across all three world models on both Cube and PushT, AIDA consistently achieves the lowest success rates and highest attack success rates, demonstrating its effectiveness over the adapted baselines. These additional experiments demonstrate that the attack performance is not specific to the original evaluation cases. The above experiments have all been added to section A.7.
Response to reviewer fC8v regarding to Weakness 2, Questions 3 and 4 (Part 3/3)
Impact of Clean Re-observation
"W2: The results do not establish that action discriminability remains impaired after the controller receives fresh observations. Equation (8) and Figure 6 concern predictions within a latent rollout. Figure 5 shows that task performance remains affected under different replanning settings, but this could also reflect damage caused by the initial incorrect actions. Both effects matter, but they are different explanations. The paper should distinguish them and either add evidence after clean observations are reintroduced or narrow the persistence claim. " and "Q3: After a fresh clean observation enters the controller, does the model still produce less distinguishable predictions across actions, or is the remaining performance loss mainly due to the earlier actions?"
Thank you for this important distinction. In the official planner, fresh clean observations are reintroduced at each replanning stage. Therefore, the performance degradation in Fig. 5 already reflects replanning with clean observations. We acknowledge that the subsequent failures are primarily due to the consequences of earlier incorrect actions. To further investigate this distinction, we compare the official planner with a variant that directly carries over the predicted latent from the previous stage instead.
| Setting | Cube SR (%) ↓ | ASR (%) ↑ | FGD (cm) ↑ | PushT SR (%) ↓ | ASR (%) ↑ | FPE ↑ | FAE (°) ↑ |
|---|---|---|---|---|---|---|---|
| Clean re-observation | 25.5 | 70.48 | 14.07 | 9.0 | 89.53 | 245.46 | 22.76 |
| Latent carryover | 16.0 | 80.72 | 14.78 | 0.5 | 99.42 | 291.44 | 22.94 |
Without clean re-observation, ASR increases by 10.24 percentage points on Cube and 9.88 points on PushT. Fresh clean observations partially mitigate the downstream impact of adversarial latent predictions. Follow your suggestion, we have revised the abstract, introduction, related work and experimental discussion to distinguish reduced action discriminability within attacked latent rollouts from downstream performance degradation across replanning stages. Experiment results have been included in section A.8.
Propagation Loss
"Q4: For the propagation loss, which history variables are differentiated, and at which prediction steps is the Jacobian evaluated?"
While the main text uses a single-frame formulation for simplicity, our implementation accounts for the full temporal context. We compute the joint Jacobian , differentiating the concatenated predictions at steps 1–3 with respect to the entire three-frame latent history while keeping the action embeddings fixed. All three predictions are obtained in a single forward pass. The spectral norm is estimated at the current perturbed history in each attack iteration, and the penalties are averaged across sampled action sequences. Only the initial observation is perturbed, while the remaining two history frames are kept unchanged during optimization. Related explanation has been added to section A.9 in the revised paper.
Response to reviewer fC8v regarding to Weakness 1, Questions 1 (Part 1/3)
We thank the reviewer for the insightful comments.
Information Needed for Attacks
Thank you for highlighting this ambiguity. We clarify that our main experiments adopt an oracle-assisted white-box setting, where the attacker is assumed to have access to a short future trajectory following the target frame, in addition to the model parameters and gradients. The expert actions guide action sampling, while future clean observations provide reference latents and reduce the influence of the world model's prediction errors. We have explicitly stated these additional assumptions in the revised threat model. To evaluate AIDA without future information, we additionally test an oracle-free variant that replaces expert actions with randomly sampled actions and future clean observations with model-generated rollout contex. This variant requires only current frame and achieves 29.8%/12% SR on Cube/PushT, compared with 24.2%/8.4% in the main setting. These results shown in the table below demonstrate that AIDA remains effective without access to future trajectories or expert actions. We have included these results and clearly distinguish the two settings in section 3.1 and 4.3.