Papers on building RL env from video via 3D GS and improving VLM via MetaVQA were accepted to CVPR 2025.