Safety
Sensitivity Shaping for Latent Modeling
arXiv:2606.14585v2 Announce Type: replace-cross Abstract: Generative dynamics models enable planning in challenging systems, but safe deployment requires detecting policy-induced out-of-distribution (
arXiv:2606.14585v2 Announce Type: replace-cross Abstract: Generative dynamics models enable planning in challenging systems, but safe deployment requires detecting policy-induced out-of-distribution (OOD) transitions. Existing methods typically treat learned dynamics as fixed and rely on post hoc support surrogates for OOD detection. This overlooks a critical failure mode: learned dynamics that are insensitive to control changes can map unsupported controls to latent predictions resembling demonstrated transitions, suppressing OOD signals despite large prediction errors. We introduce support-conditioned control-sensitivity regularization to preserve control-induced variation by promoting local responsiveness in well-supported training regions. Experiments in vision-based obstacle avoidance, manipulation, and real-robot navigation demonstrate improved OOD detection and safer closed-loop planning.
Related
- PH-Dreamer: A Physics-Driven World Model via Port-Hamiltonian Generative Dynamics
- PILD: Physics-Informed Learning via Diffusion
- Imagine-then-Plan: Agent Learning from Adaptive Lookahead with World Models
- Olaf-World: Orienting Latent Actions for Video World Modeling
- Transferable Reinforcement Learning via Probabilistic Latent Embeddings and Dynamic Policy Adaptation for Sim-to-Real Deployment
Source: arXiv cs.AI | 2026-10-07