Realtime AI News
OPPO: Omni-Perception Policy Optimization Framework for Multimodal Emotion Reasoning
A new arXiv paper proposes OPPO, a reinforcement learning framework that explicitly optimizes multimodal perception for emotion reasoning, addressing the underutilization of multimodal cues and hallucination in current Omni-MLLMs.
A paper titled 'Omni-Perception Policy Optimization for Multimodal Emotion Reasoning' has been published on arXiv. The research finds that current emotion-oriented Omni-MLLMs still lack reliable omni-modal perception: they underutilize multimodal cues in their reasoning trajectories and exhibit unfaithful behavior, often hallucinating modality-specific statements from other modalities.
Building on these insights, the authors propose OPPO, a reinforcement learning framework that explicitly optimizes multimodal perception. An Omni-Perception Reward Model is introduced to guide the learning process.
The source is arXiv cs.AI (ID 2606.25325), published on June 25, 2026.
Why it matters
OPPO directly addresses the perception reliability and faithfulness issues in multimodal emotion AI, offering a new paradigm for building more trustworthy affective interaction systems.