Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Control
World Action Models (WAMs) enable semantically- and physically-informed control but are brittle under distribution shift. In this work, we use mechanistic interpretability to study how robustness-relevant perturbations are represented in WAM activation space. Comparing activations across successful and unsuccessful rollouts, we find some WAM architectures exhibit low-dimensional linear separability for robustness-critical features, while others do not. This motivates the use of contrastive activation directions for training-free WAM steering. We also show that local linearity in WAM activation
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- FuzzySimilar title/name (fuzzy) · 87%lllyasviel/ControlNet →
“Fuzzy title match (0.94): “Steering Robustness into World Action Models via Mechanistic” ≈ “lllyasviel/ControlNet””
- FuzzySimilar title/name (fuzzy) · 87%lllyasviel/ControlNet-v1-1 →
“Fuzzy title match (0.94): “Steering Robustness into World Action Models via Mechanistic” ≈ “lllyasviel/ControlNet-v1-1””
- PossiblePossibly related (embedding) · 45%Gradient-based Planning for World Models at Longer Horizons →
- FuzzySimilar title/name (fuzzy) · 87%google-deepmind/dm_control →
“Fuzzy title match (0.94): “Steering Robustness into World Action Models via Mechanistic” ≈ “google-deepmind/dm_control””
- FuzzyOverlapping authors or contributors · 62%zhayujie/CowAgent →
“Shared author/contributor keys: chan”
- FuzzySimilar title/name (fuzzy) · 59%builderz-labs/mission-control →
“Fuzzy title match (0.73): “Steering Robustness into World Action Models via Mechanistic” ≈ “builderz-labs/mission-control””
- LinkedLinked via arxiv author · 85%Jihoon Hong →
“Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Control”
- LinkedLinked via arxiv author · 85%Julian Skifstad →
“Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Control”