MiMo-Embodied: X-Embodied Foundation Model Technical Report
MiMo-Embodied: X-Embodied Foundation Model Technical Report [53.3] 私たちはMiMo-Embodiedをオープンソースとして公開しました。 MiMo-Embodiedはタスクプランニング、アフォーマンス予測、空間理解において17のAIベンチマークにまたがる新たな記録を設定している。 これらのタスク全体で、MiMo-Embodiedは既存のオープンソース、クローズドソース、および特別なベースラインを著しく上回っている。 論文参考訳(メタデータ) (Thu, 20 Nov 2025 16:34:55 GMT)
「This report introduces MiMo-Embodied, a pioneering cross-embodied vision-language model that achieves state-of-the-art performance in both autonomous driving and embodied AI tasks. As the first open-source VLM integrating these two critical domains, MiMo-Embodied significantly enhances understanding and reasoning in dynamic physical environments. Extensive evaluations across 29 benchmarks show that MiMo-Embodied achieves superior performance in both embodied and autonomous driving tasks, significantly outperforming existing open-source and closed-source general VLMs, as well as specialized VLMs for a single domain」とEmbdiedなタスクで高性能を主張する公開モデル。ドメインによる得意不得意はあるだろうが、商用モデルの性能も超えている。