MoST: Multi-modality Scene Tokenization for Motion Prediction | IEEE Conference Publication | IEEE Xplore