Embodied-R1: Reinforced Embodied Reasoning for General Robotic Manipulation

2025/08/19 by Yifu Yuan, Yuan, Yifu, Yao‐Ting Huang +15 · 15 citations
Computer Science · Engineering · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Reinforcement Learning in Robotics #Robot Manipulation and Learning #Robotics (cs.RO)

paper · pdf · doi:10.48550/arxiv.2508.13998

openalex publication_date 2025/08/19 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Generalization in embodied AI is hindered by the "seeing-to-doing gap," which stems from data scarcity and embodiment heterogeneity. To address this, we pioneer "pointing" as a unified, embodiment-agnostic intermediate representation, defining four core embodied pointing abilities that bridge high-level vision-language comprehension with low-level action primitives. We introduce Embodied-R1, a 3B Vision-Language Model (VLM) specifically designed for embodied reasoning and pointing. We use a wide range of embodied and general visual reasoning datasets as sources to construct a large-scale dataset, Embodied-Points-200K, which supports key embodied pointing capabilities. We then train Embodied-R1 using a two-stage Reinforced Fine-tuning (RFT) curriculum with a specialized multi-task reward design. Embodied-R1 achieves state-of-the-art performance on 11 embodied spatial and pointing benchmarks. Critically, it demonstrates robust zero-shot generalization by achieving a 56.2% success rate in the SIMPLEREnv and 87.5% across 8 real-world XArm tasks without any task-specific fine-tuning, representing a 62% improvement over strong baselines. Furthermore, the model exhibits high robustness against diverse visual disturbances. Our work shows that a pointing-centric representation, combined with an RFT training paradigm, offers an effective and generalizable pathway to closing the perception-action gap in robotics.

Citations

• MolmoAct: Action Reasoning Models that can Reason in Space
• ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning
• Where, What, Why: Towards Explainable Driver Attention Prediction
• What Can RL Bring to VLA Generalization? An Empirical Study
• VLA-RL: Towards Masterful and General Robotic Manipulation with Scalable Reinforcement Learning
• A0: An Affordance-Aware Hierarchical Model for General Robotic Manipulation
• VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning
• DAPO: An Open-Source LLM Reinforcement Learning System at Scale
• RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete
• Qwen2.5-VL Technical Report
• SoFar: Language-Grounded Orientation Bridges Spatial Reasoning and Object Manipulation
• HAMSTER: Hierarchical Action Models For Open-World Robot Manipulation
• ConRFT: A Reinforced Fine-tuning Method for VLA Models via Consistency Policy
• Action-Free Reasoning for Policy Generalization
• SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training
• SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Model
• Kimi k1.5: Scaling Reinforcement Learning with LLMs
• DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning
• SpatialCoT: Advancing Spatial Reasoning through Coordinate Alignment and Chain-of-Thought for Embodied Task Planning
• What Matters in Building Vision-Language-Action Models for Generalist Robots
• Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces
• TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policies
• SAT: Dynamic Spatial Aptitude Training for Multimodal Language Models
• RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics
• RT-Affordance: Affordances are Versatile Intermediate Representations for Robot Manipulation
• GPT-4o System Card
• Data Scaling Laws in Imitation Learning for Robotic Manipulation
• ManiSkill3: GPU Parallelized Robotics Simulation and Rendering for Generalizable Embodied AI
• Reasoning Paths with Reference Objects Elicit Quantitative Spatial Reasoning in Large Vision-Language Models
• ReKep: Spatio-Temporal Reasoning of Relational Keypoint Constraints for Robotic Manipulation
• Flow as the Cross-Domain Manipulation Interface
• Robotic Control via Embodied Chain-of-Thought Reasoning
• LLaRA: Supercharging Robot Learning Data for Vision-Language Policy
• SpatialBot: Precise Spatial Understanding with Vision Language Models
• RoboPoint: A Vision-Language Model for Spatial Affordance Prediction for Robotics
• OpenVLA: An Open-Source Vision-Language-Action Model
• A3VLM: Actionable Articulation-Aware Vision Language Model
• EmbSpatial-Bench: Benchmarking Spatial Understanding for Embodied Tasks with Large Vision-Language Models
• TopViewRS: Vision-Language Models as Top-View Spatial Reasoners
• SpatialRGPT: Grounded Spatial Reasoning in Vision Language Models
• Octo: An Open-Source Generalist Robot Policy
• Evaluating Real-World Robot Manipulation Policies in Simulation
• BLINK: Multimodal Large Language Models Can See but Not Perceive
• ManipVQA: Injecting Robotic Affordance and Physically Grounded Information into Multi-Modal Large Language Models
• MOKA: Open-World Robotic Manipulation through Mark-Based Visual Prompting
• The All-Seeing Project V2: Towards General Relation Comprehension of the Open World
• BioXP-0.5B: Explainable Medical-AI via RL-GRPO
• Grounded SAM: Assembling Open-World Models for Diverse Visual Tasks
• SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities
• General Flow as Foundation Affordance for Scalable Robot Learning
• Any-point Trajectory Modeling for Policy Learning
• Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks
• RT-Trajectory: Robotic Task Generalization via Hindsight Trajectory Sketches
• What's "up" with vision-language models? Investigating their struggle with spatial reasoning
• Open X-Embodiment: Robotic Learning Datasets and RT-X Models
• SWE-bench: Can Language Models Resolve Real-World GitHub Issues?
• BridgeData V2: A Dataset for Robot Learning at Scale
• HANDAL: A Dataset of Real-World Manipulable Object Categories with Pose Annotations, Affordances, and Reconstructions
• VL-Grasp: a 6-Dof Interactive Grasp Policy for Language-Oriented Objects in Cluttered Indoor Scenes
• VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models
• Objaverse-XL: A Universe of 10M+ 3D Objects
• LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning
• Micrograph segmentations for DDEVD
• Segment Anything
• PoseCNN: A Convolutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes
• Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models
• 3D-LLM: Injecting the 3D World into Large Language Models

Cited by

• Universal Dexterous Functional Grasping via Demonstration-Editing Reinforcement Learning
• TraceGen: World Modeling in 3D Trace Space Enables Learning from Cross-Embodiment Videos
• LocateAnything3D: Vision-Language 3D Detection with Chain-of-Sight
• iFlyBot-VLM Technical Report
• Pelican-VL 1.0: A Foundation Brain Model for Embodied Intelligence
• Vlaser: Vision-Language-Action Model with Synergistic Embodied Reasoning
• A Survey on Agentic Multimodal Large Language Models
• More than A Point: Capturing Uncertainty with Adaptive Affordance Heatmaps for Spatial Grounding in Robotic Tasks
• PhysToolBench: Benchmarking Physical Tool Understanding for MLLMs
• Reinforced Embodied Planning with Verifiable Reward for Real-World Robotic Manipulation
• Pure Vision Language Action (VLA) Models: A Comprehensive Survey
• RoboSeek: You Need to Interact with Your Objects
• Embodied Arena: A Comprehensive, Unified, and Evolving Evaluation Platform for Embodied AI
• From Failures to Supervision: DynamicEnvPlan for Robust Long-Horizon Embodied Planning
• Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models

Related

• Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models
• From Seeing to Doing: Bridging Reasoning and Decision for Robotic Manipulation
• Unified Embodied VLM Reasoning with Robotic Action via Autoregressive Discretized Pre-training
• Point-It-Out: Benchmarking Embodied Reasoning for Vision Language Models in Multi-Stage Visual Grounding
• Mind to Hand: Purposeful Robotic Control via Embodied Reasoning