VLAs with Long and Short-Term Memory

Multi-Scale Embodied Memory (MEM) gives our models both long-term and short-term memory, enabling complex tasks longer than ten minutes.

VLAs with Long and Short-Term Memory VLAs with Long and Short-Term Memory Published March 3, 2026 Email research@physicalintelligence.company Marcel Torne, Karl Pertsch, Homer Walke, Kyle Vedder, Suraj Nair, Brian Ichter, Allen Ren, Haohuan Wang, Jiaming Tang, Kyle Stachowicz, Karan Dhabalia, Michael Equi, Quan Vuong, Jost Tobias Springenberg, Sergey Levine, Chelsea Finn, Danny Driess Paper MEM .pdf Loading… Current robotic foundation models can follow commands like "wash the frying pan" , "fold the laundry" , or "make a peanut butter sandwich" , performing intricate and complex motor skills.

saved by

• Jennifer Zhao
• Timothy Kostolansky

related reading

• A VLA with Open-World Generalizationpi.website
• MemER: Scaling Up Memory for Robot Control via Experience Retrievalarxiv.org
• LLM Powered Autonomous Agents | Lil'Loglilianweng.github.io
• How can we develop transformative tools for thought?numinous.productions
• Augmenting Long-term Memoryaugmentingcognition.com
• Augmenting Long-term Memoryaugmentingcognition.com
• Adding Long-Term Memory to a chatbot with Mem0 | by Rajesh Nayak | Mediummedium.com
• Mem0 - AI Memory Layer for your Agents & Apps | Persistent Contextmem0.ai
• What are the differences between long-term, short-term, and working memory? - PMCncbi.nlm.nih.gov
• [2501.00663] Titans: Learning to Memorize at Test Timearxiv.org
• [2605.15156] MeMo: Memory as a Modelarxiv.org
• When Models Manipulate Manifolds: The Geometry of a Counting Tasktransformer-circuits.pub