Latent Dynamics · 2026-11-13
Does Your World Model Know Actions Have Consequences?
Ayush Gaggar, Jared Berry, Todd D. Murphey
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors argue current embodied world-model evaluation misses how well action-conditioned dynamics predict counterfactual consequences. They introduce an Action Perturbation (AP) sensitivity metric comparing predicted vs true impulse responses under forced action perturbations. Using frozen TD-MPC2 components and finetuned dynamics heads trained with different data-collection regimes, they show noise-injected heads improve action sensitivity while workspace raster performs worst.
Why it matters
The paper targets evaluation “fidelity of modeling consequences of counterfactual actions,” which the authors say is not captured by reconstruction or task success metrics. By using sensitivity analysis over action perturbations and defining AP, the work provides a way to audit an embodied world model’s dynamics predictor for planning/decision transparency via counterfactual auditing, and highlights tradeoffs between state error and sensitivity accuracy.
Method
- Model forced response to action perturbations using sensitivity analysis; define Action Perturbation (AP) as a normalized discrepancy between predicted and true impulse responses under counterfactual action perturbations.
- Freeze TD-MPC2 checkpoints for non-dynamics components; finetune only the dynamics head on 100k interactions from three data-collection regimes (workspace raster, imitation learning, noise-injected).
- Evaluate predicted state error and AP over 50 fixed, on-policy trajectories (push and peg tasks) using action perturbations applied at the first step and open-loop rollout for a horizon H.
Limitation
The AP metric “requires precise resetting of the environment to compare the effect of various action perturbations from the same initial state, which is inherently limited to simulation.”
Abstract (from arXiv)
Embodied world models (WMs) are implicit learning systems that predict the next future latent state directly from observations and actions. Unique from other end-to-end learning systems, the dynamics predictor of WMs can simulate the environment's response to alternative actions without physical deployment nor pre-specifying states of the dynamical system. Current metrics focus on the quality of model outputs, e.g., clarity of reconstructions or task success, but fail to assess the fidelity of modeling consequences of counterfactual actions. Inspecting the forced response to action perturbations through sensitivity analysis is one way to audit how well WMs can predict the consequences of counterfactual actions. Deploying WMs on hardware will require new evaluation frameworks to analyze the tradeoffs between state error and accurate sensitivity prediction.
Related papers
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First announced by arXiv: 2026-11-13 · Indexed by Pub-AI: 2026-11-13 UTC
Does Your World Model Know Actions Have Consequences?
First announced by arXiv: 2026-11-13 · Original record ↗
- arxiv
- 2611.15048
Explicit source versions (1)
- v1 · Date not supplied
- Preprint available
Source date: 2026-11-13 · First detected: 2026-11-13 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.15048v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv
Embodied world models (WMs) are implicit learning systems that predict the next future latent state directly from observations and actions. Unique from other end-to-end learning systems, the dynamics predictor of WMs can simulate the environment's response to alternative actions without physical deployment nor pre-specifying states of the dynamical system. Current metrics focus on the quality of model outputs, e.g., clarity of reconstructions or task success, but fail to assess the fidelity of modeling consequences of counterfactual actions. Inspecting the forced response to action perturbations through sensitivity analysis is one way to audit how well WMs can predict the consequences of counterfactual actions. Deploying WMs on hardware will require new evaluation frameworks to analyze the tradeoffs between state error and accurate sensitivity prediction.
