UniLION: Towards Unified Autonomous Driving Model with Linear Group RNNs
UniLION: Towards Unified Autonomous Driving Model with Linear Group RNNs [115.9] UniLIONは、大規模なLiDAR点雲、高解像度のマルチビュー画像、さらには時間的シーケンスを効率的に処理する。 UniLIONは、幅広いコアタスクにわたって、競争力と最先端のパフォーマンスを一貫して提供します。 論文参考訳（メタデータ） (Mon, 03 Nov 2025 17:24:19 GMT)
「We propose UniLION, a unified model that achieves both latent temporal fusion and multimodal fusion in UniLION backbone by the linear group RNN, generating the unified BEV features that serve all autonomous driving tasks, including perception, prediction, and planning.」とRNNベースのマルチモーダルモデルの提案。「Unified Heterogeneous Inputs: Leveraging the superior long-range modeling capability and linear computational complexity of linear group RNNs, UniLION integrates multi-view images, LiDAR point clouds, and temporal information into a unified 3D backbone through direct token concatenation, eliminating hand-crafted fusion modules and providing a more elegant, scalable solution.」ととてもマルチモーダル。