TSTM: Temporal Segmentation for Task-relevant Mask in Visual Reinforcement Learning Generalization

Описание: This video presents our CVPR 2026 paper “TSTM: Temporal Segmentation for Task-relevant Mask in Visual Reinforcement Learning Generalization.”
We propose TSTM, a temporal segmentation framework that leverages observation sequences to produce more reliable task-relevant masks and improve policy generalization under visual distractions.