RoboMemory: A Brain-inspired Multi-memory Agentic Framework for Interactive Environmental Learning in Physical Embodied Systems

2025/08/02 by Lei Mo, Lei, Mingcong, Han Cai +29 · 7 citations
Computer Science · Engineering · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Modular Robots and Swarm Intelligence #Multi-Agent Systems and Negotiation #Reinforcement Learning in Robotics #Robotics (cs.RO)

paper · pdf · doi:10.48550/arxiv.2508.01415

openalex publication_date 2025/08/02 · openalex created_date 2025/10/09 · openalex updated_date 2026/07/28

Abstract

Embodied intelligence aims to enable robots to learn, reason, and generalize robustly across complex real-world environments. However, existing approaches often struggle with partial observability, fragmented spatial reasoning, and inefficient integration of heterogeneous memories, limiting their capacity for long-horizon adaptation. To address this, we introduce RoboMemory, a brain-inspired framework that unifies Spatial, Temporal, Episodic, and Semantic memory within a parallelized architecture for efficient long-horizon planning and interactive learning. Its core innovations are a dynamic spatial knowledge graph for scalable, consistent memory updates and a closed-loop planner with a critic module for adaptive decision-making. Extensive experiments on EmbodiedBench show that RoboMemory, instantiated with Qwen2.5-VL-72B-Ins, improves the average success rate by 26.5% over its strong baseline and even surpasses the closed-source SOTA, Claude-3.5-Sonnet. Real-world trials further confirm its capability for cumulative learning, with performance consistently improving over repeated tasks. Our results position RoboMemory as a scalable foundation for memory-augmented embodied agents, bridging insights from cognitive neuroscience with practical robotic autonomy.

Citations

• NavA³: Understanding Any Instruction, Navigating Anywhere, Finding Anything
• RoboBrain 2.0 Technical Report
• Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models
• Agentic Robot: A Brain-Inspired Framework for Vision-Language-Action Models in Embodied Agents
• RoboOS: A Hierarchical Embodied Framework for Cross-Embodiment and Multi-Agent Collaboration
• LLM-Empowered Embodied Agent for Memory-Augmented Task Planning in Household Robotics
• Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory
• InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
• Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems
• GR00T N1: An Open Foundation Model for Generalist Humanoid Robots
• Being-0: A Humanoid Robotic Agent with Vision-Language Models and Modular Skills
• NeSyC: A Neuro-symbolic Continual Learner For Complex Embodied Tasks In Open Domains
• CLEA: Closed-Loop Embodied Agent for Enhancing Task Execution in Dynamic Environments
• Hi Robot: Open-Ended Instruction Following with Hierarchical Vision-Language-Action Models
• Qwen2.5-VL Technical Report
• EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents
• Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
• 3D-Mem: 3D Scene Memory for Embodied Exploration and Reasoning
• π₀: A Vision-Language-Action Flow Model for General Robot Control
• GPT-4o System Card
• Agent S: An Open Agentic Framework that Uses Computers Like a Human
• Open Scene Graphs for Open-World Object-Goal Navigation
• OpenVLA: An Open-Source Vision-Language-Action Model
• AutoManual: Constructing Instruction Manuals by LLM Agents via Interactive Environmental Learning
• HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models
• Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context
• Cradle: Empowering Foundation Agents Towards General Computer Control
• Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation
• Retrieval-Augmented Generation for Large Language Models: A Survey
• Toward General-Purpose Robots via Foundation Models: A Survey and Meta-Analysis
• GOAT: GO to Any Thing
• MemGPT: Towards LLMs as Operating Systems
• ConceptGraphs: Open-Vocabulary 3D Scene Graphs for Perception and Planning
• ExpeL: LLM Agents Are Experiential Learners
• Building Cooperative Embodied Agents Modularly with Large Language Models
• SwiftSage: A Generative Agent with Fast and Slow Thinking for Complex Interactive Tasks
• Voyager: An Open-Ended Embodied Agent with Large Language Models
• Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
• Generative Agents: Interactive Simulacra of Human Behavior
• Reflexion: Language Agents with Verbal Reinforcement Learning
• Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
• Diffusion policy: Visuomotor policy learning via action diffusion
• LLM-Planner: Few-Shot Grounded Planning for Embodied Agents with Large Language Models
• Distributing Accountability, Not Capability: Phase Separation and the LLM Workflow Quadrant in Autonomous AI Agent Architectures
• LiveBench: A Challenging, Contamination-Limited LLM Benchmark

Cited by

• Physical AI Governance: From Theory to Practice Across Life Cycle
• Learning Evolving Latent Strategies for Multi-Agent Language Systems without Model Fine-Tuning
• VLA²: Empowering Vision-Language-Action Models with an Agentic Framework for Unseen Concept Manipulation
• From Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms
• BridgeVLA++: A Data-Efficient, Generalizable, and Memory-Augmented Vision-Language-Action Framework for 3D Manipulation
• In-Context VLA: Endowing Vision-Language-Action Models with Language via In-Context Post-Training and Agentic Tool Use
• SkillMemo: Expert-guided Skill Memory Framework for Compositional Embodied Manipulation

Related

• Robobench: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models as Embodied Brain
• LLM as A Robotic Brain: Unifying Egocentric Memory and Control
• RoboOS: A Hierarchical Embodied Framework for Cross-Embodiment and Multi-Agent Collaboration
• RoboOS-NeXT: A Unified Memory-based Framework for Lifelong, Scalable, and Robust Multi-Robot Collaboration
• KARMA: Augmenting Embodied AI Agents with Long-and-short Term Memory Systems