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

ICLR 2026 Conference Submission19655 Authors

Keywords: Embodied AI, Agentic AI, Interactive Environmental Learning, Brain-like Memory

TL;DR: RoboMemory, a four-module embodied framework, advances real-world task success via lifelong learning, resolving dynamic continuous learning, multi-module memory latency, task correlations for long-term improvement, and closed-loop planning loops.

Abstract: Embodied agents face persistent challenges in real-world environments, including partial observability, limited spatial reasoning, and high-latency multi-memory integration. We present RoboMemory, a brain-inspired framework that unifies Spatial, Temporal, Episodic, and Semantic memory under a parallelized architecture for efficient long-horizon planning and interactive environmental learning. A dynamic spatial knowledge graph (KG) ensures scalable and consistent memory updates, while a closed-loop planner with a critic module supports adaptive decision-making in dynamic settings. Experiments on EmbodiedBench show that RoboMemory, built on Qwen2.5-VL-72B-Ins, improves average success rates by 25% over its baseline and exceeds the closed-source state-of-the-art (SOTA) Gemini-1.5-Pro by 3%. Real-world trials further confirm its capacity for cumulative learning, with performance improving across repeated tasks. These results highlight RoboMemory as a scalable foundation for memory-augmented embodied intelligence, bridging the gap between cognitive neuroscience and robotic autonomy.

Primary Area: applications to robotics, autonomy, planning

Submission Number: 19655