Embodied-Reasoner: Synergizing Visual Search, Reasoning, and Action for Embodied Interactive Tasks

TL;DR

The paper tackles the challenge of extending deep-thinking reasoning to embodied interactive tasks. It introduces Embodied-Reasoner, a model that interleaves Observation with diverse Thought processes before actions, powered by a data engine that synthesizes 9.3k Observation-Thought-Action trajectories and a three-stage training pipeline (imitation, rejection sampling, reflection tuning). Evaluations on AI2-THOR across Search, Manipulation, Transport, and Composite tasks show substantial gains over state-of-the-art visual-language models, with the largest improvements on long-horizon, complex tasks and real-world generalization. This work provides a practical framework for translating long-form reasoning into embodied agents through structured data generation and iterative self-improvement, highlighting significant potential for robust, interactive intelligent systems.

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.

Figures (18)

• Figure 1: We design an embodied interactive task: searching for objects in an unknown room. Then we propose Embodied-Reasoner, which presents spontaneous reasoning and interaction ability. Before each action, it generates diverse thoughts, e.g., self-reflection or spatial reasoning, forming an image-text interleaved trajectory. It shows consistent reasoning and efficient search behaviors, whereas OpenAI o3-mini often exhibits repetitive searches and logical inconsistencies with higher failure rates.
• Figure 2: Embodied-Reasoner exhibits spontaneous thinking behaviors, e.g., analyzing environmental states (#1,3), reflecting on missed details (#4), reasoning based on the latest observations (#5), and recalling cues for efficient planning (#9). These thoughts remain coherent and logically consistent despite spanning multiple rounds. In contrast, general VLMs lacking thinking abilities struggle with long-horizon interactive tasks and produce unreasonable actions, e.g., forget tasks or repetitive searching.
• Figure 3: Left: Data Engine for $<$Instruction, Interactive Trajectory$>$ synthesis. First, we synthesize instructions from task templates, and build an affiliation graph from scene's meta-data. It enables us to derive key actions needed for task. We add exploratory actions and insert thinking thoughts between observation and actions. Right: Three-stage training recipe. ①We fine-tune on synthesized trajectory to develop interaction skills. ②We sample multiple trajectories on novel tasks and evaluate their correctness. The successful ones are used for developing its exploring abilities. ③We continue to sample trajectories using updated model, injecting anomalous states and reflective thoughts in successful cases and correcting errors in failed ones. This self-correction training yields Embodied-Reasoner.
• Figure 4: We analyze the frequency of five types of thoughts and their flexible transition relationships in all trajectories.
• Figure 5: Relations between task length and success rate, and output token number. As task complexity increases, our model generates more reasoning tokens to maintain high success rates.