TGPO：通过可验证奖励强化学习激励第一人称视频时序感知
阅读原文· machinelearning.apple.com多模态大语言模型（MLLM）在第一人称视频理解中缺乏时序感知，常依赖空间捷径。为此，研究者提出 Temporal Global Policy Optimization（TGPO），一种基于可验证奖励的强化学习算法。TGPO 通过对比模型在时序有序与打乱帧上的输出，生成全局归一化奖励信号，明确奖励时序连贯推理。TGPO 可集成 GRPO 和 GSPO，支持冷启动 RL 训练，抑制 MLLM 的空间捷径行为。在五个第一人称视频基准上，TGPO 一致提升时序定位与因果连贯性，优于此前基于 RL 的视频推理方法。
research area Computer Vision, research area Methods and Algorithms
content type paperpublished July 2026
Incentivizing Temporal-Awareness in Egocentric Video Understanding Models
Authors Zhiyang Xu†, Tian Qin‡, Bowen Jin§, Zhengfeng Lai¶, Meng Cao, Lifu Huang¶, Peng Zhang
Multimodal large language models (MLLMs) have recently shown strong performance in visual understanding, yet they often lack temporal awareness, particularly in egocentric settings where reasoning depends on the correct ordering and evolution of events. This deficiency stems in part from training objectives that fail to explicitly reward temporal reasoning and instead rely on frame-level spatial shortcuts. To address this limitation, we propose Temporal Global Policy Optimization (TGPO), a reinforcement learning with verifiable rewards (RLVR) algorithm designed to incentivize temporal awareness in MLLMs. TGPO contrasts model outputs generated from temporally ordered versus shuffled video frames to derive calibrated, globally normalized reward signals that explicitly favor temporally coherent reasoning. Integrated with GRPO and GSPO, TGPO supports cold-start RL training and effectively suppresses spatial shortcut behaviors learned by existing MLLMs. Experiments across five egocentric video benchmarks demonstrate that TGPO consistently improves temporal grounding and causal coherence, outperforming prior RL-based video reasoning approaches. Our results suggest that TGPO offers a simple and scalable pathway toward temporally robust MLLMs for egocentric video understanding.