Embodied-Reasoner: Synergizing Visual Search, Reasoning, and Action for Embodied Interactive Tasks
🤖 AI Summary
Embodied intelligence faces challenges in jointly coordinating visual search, temporal reasoning, and action planning. Method: We propose the first deep-thinking model tailored for embodied interaction. Our approach introduces a high-quality 9.3K Observation-Thought-Action trajectory dataset and a three-stage training paradigm—imitation learning, rejection-sampling-based self-exploration, and reflection-based fine-tuning—to jointly optimize spatial perception, temporal modeling, and historical reflection. The architecture adopts an o1-style multimodal design, integrating visual encoding, chain-of-thought reasoning, action policy generation, and reflective self-correction. Results: Our model achieves significant improvements over OpenAI o1, o3-mini, and Claude-3.7 on embodied search tasks (+9%, +24%, and +13%, respectively), markedly reduces repeated searches and logical inconsistencies, and demonstrates superior performance on long-horizon tasks.
Application Category
📝 Abstract
Recent advances in deep thinking models have demonstrated remarkable reasoning capabilities on mathematical and coding tasks. However, their effectiveness in embodied domains which require continuous interaction with environments through image action interleaved trajectories remains largely -unexplored. We present Embodied Reasoner, a model that extends o1 style reasoning to interactive embodied search tasks. Unlike mathematical reasoning that relies primarily on logical deduction, embodied scenarios demand spatial understanding, temporal reasoning, and ongoing self-reflection based on interaction history. To address these challenges, we synthesize 9.3k coherent Observation-Thought-Action trajectories containing 64k interactive images and 90k diverse thinking processes (analysis, spatial reasoning, reflection, planning, and verification). We develop a three-stage training pipeline that progressively enhances the model's capabilities through imitation learning, self-exploration via rejection sampling, and self-correction through reflection tuning. The evaluation shows that our model significantly outperforms those advanced visual reasoning models, e.g., it exceeds OpenAI o1, o3-mini, and Claude-3.7 by +9%, 24%, and +13%. Analysis reveals our model exhibits fewer repeated searches and logical inconsistencies, with particular advantages in complex long-horizon tasks. Real-world environments also show our superiority while exhibiting fewer repeated searches and logical inconsistency cases.
Problem
Research questions and friction points this paper is trying to address.
Extends reasoning to interactive embodied search tasks
Requires spatial understanding and temporal reasoning
Addresses repeated searches and logical inconsistencies
Innovation
Methods, ideas, or system contributions that make the work stand out.
Extends reasoning to interactive embodied search tasks
Uses Observation-Thought-Action trajectories for training
Three-stage training pipeline enhances model capabilities
🔎 Similar Papers
No similar papers found.
