Multimodal Filesystem Memory For Long-Horizon Embodied Agents
We present MFM, a multimodal filesystem memory architecture that keeps complementary types of agent memory.
ProofPaper ↗
Key points
- Agents operating over long horizons need an effective memory that allows them to reuse their experience throughout long interactions.
- Episodic memory stores the observations, actions, and outcomes, and spatial memory associates experiences with locations.
- MFM makes use of searchable textual records to connect these different representations to the underlying multimodal episodes.
- In the case of long-horizon Minecraft and computer-use tasks, MFM leads to improved performance with three base models.
Sources (1)
- [1]Multimodal Filesystem Memory For Long-Horizon Embodied AgentsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 09:56 PM
We present MFM, a multimodal filesystem memory architecture that keeps complementary types of agent memory.
Agents operating over long horizons need an effective memory that allows them to reuse their experience throughout long interactions.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 15, 2026StepFun Releases Step 5 Preview Open Weights: 600B MoE, 27B Active, 1M Context
- Oct 15, 2026OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
- Oct 15, 2026Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio
- Oct 15, 2026Has GPT-6-Astra Achieved Embodied Intelligence?
- Oct 15, 2026LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use Agents
- Oct 14, 2026Amp puts AI agents to work beyond code with OpenAI models
