Embodied-Reasoner Outperforms OpenAI o1 in Embodied AI Tasks

A joint team from Zhejiang University, the Chinese Academy of Sciences, and Alibaba Damo Academy has open-sourced Embodied-Reasoner, a multimodal embodied reasoning model that achieved an 80.96% task success rate in the AI2-THOR simulator. This performance surpasses OpenAI o1 (71.73%), o3-mini (56.55%), and Claude-3.7 (67.70%). The model features advanced visual search, spatial reasoning, and self-correction capabilities achieved through an innovative three-stage training pipeline including imitation learning, self-exploration, and reflection tuning.