Sensitivity Shaping for Latent Modeling

2026-06-12

AI Summary

This work addresses a critical blind spot in existing learned dynamics models: their insensitivity to control inputs in out-of-distribution (OOD) states, which can lead to unsafe behavior. To mitigate this, the authors propose a control sensitivity regularization mechanism that enhances the model’s local sensitivity to action perturbations within high-data-support regions while suppressing unreliable extrapolation in low-support regions. This approach enables effective OOD detection without requiring post-hoc processing, thereby significantly improving the safety of closed-loop planning. Experimental results demonstrate consistent improvements over baseline methods in both OOD detection accuracy and planning robustness across diverse tasks, including visual obstacle avoidance, manipulation, and real-world robotic navigation.

Abstract

Generative dynamics models enable planning in challenging robotic systems, but safe deployment requires reliably detecting policy-induced out-of-distribution (OOD) transitions. Existing methods typically treat the learned dynamics as fixed and attach post hoc support surrogates. We show that these surrogates can fail when the dynamics are locally insensitive to critical action choices: unsupported control actions may produce latent predictions that resemble demonstrated transitions, suppressing OOD signals despite large true predictive errors. To address this, we introduce support-conditioned control-sensitivity regularization, which promotes sensitive local response to control input changes in learned dynamics in high-support training regions. This preserves control-induced variation while limiting unstable extrapolation due to weak empirical support. Experiments in vision-based obstacle avoidance, manipulation, and real-robot navigation show improved OOD detection and safer closed-loop planning.

Problem

• out-of-distribution detection
• generative dynamics models
• control sensitivity
• robotic planning
• latent modeling

Innovation

• control-sensitivity regularization
• out-of-distribution detection
• latent dynamics modeling
• safe reinforcement learning
• support-conditioned learning