The field of computer vision and natural language processing is moving towards more sophisticated and interpretable models, with a focus on multimodal reasoning and segmentation. Recent developments have shown that integrating reinforcement learning and chain-of-thought reasoning can significantly improve the performance and generalizability of models in tasks such as human-object interaction detection, video reasoning segmentation, and image annotation. These advances have the potential to enable more accurate and robust models for applications such as AR/VR, robotics, and human-computer interaction. Noteworthy papers in this area include HOID-R1, which achieves state-of-the-art performance on HOI detection benchmarks, and Veason-R1, which surpasses prior art in video reasoning segmentation. Additionally, RISE and LENS propose innovative frameworks for enhancing vision-language models with self-supervised reasoning and unified reinforced reasoning, respectively.