Safety
Calibration-risk routing for controlled world-model adaptation
arXiv:2610.01001v1 Announce Type: new Abstract: Model-based reinforcement learning (MBRL) can exploit simulated experience, but a simulator-to-target shift creates a model-selection problem: correctin
arXiv:2610.01001v1 Announce Type: new Abstract: Model-based reinforcement learning (MBRL) can exploit simulated experience, but a simulator-to-target shift creates a model-selection problem: correcting the simulator and fitting the target directly can each fail under limited target data. We introduce the Model-Corrected World Model (MC-WM), which separates initial target data into disjoint fit, selection, and calibration partitions and deploys the family with lower standardized calibration risk. A learned confidence signal and deterministic validity predicates weight one-step imagined policy updates without rewriting physical rewards. We evaluate 540 unique reported run cells across three controlled Multi-Joint dynamics with Contact (MuJoCo) shifts; one exact-routing cell was repeated after a pre-deployment artifact gate, giving 541 completed executions.
Related
- Post-Hoc Robustness for Model-Based Reinforcement Learning
- Sampling-Based Safe Reinforcement Learning
- CROP: Conservative Reward for Model-based Offline Policy Optimization
- Theoretical Foundations and Effective Algorithms for Policy-Aware Simulator Learning
Source: arXiv cs.AI | 2026-10-02