Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action | IEEE Conference Publication | IEEE Xplore