Stanford's MessyMem Gives Robots Persistent Cross-Scene Memory

Stanford's MessyMem, accepted by CoRL 2026, gives mobile manipulation robots persistent learn-from-experience memory. Ablations show all three components matter, with combined accuracy reaching 84% and 99%; in long-horizon tests it achieved 80% progress across 25 tasks in 3 hours.

2026-09-17 ~ 2026-09-17 · 3 related posts

• MessyMem (CoRL 2026): persistent memory for robots via 3D scene graphs and VLM analysis — leto__jean · 2026-09-17
• Stanford's MessyMem gives robots long-horizon memory: 80% progress over 25 tasks spanning 3+ hours — leto__jean · 2026-09-17
• MessyMem ablations: scene graph + interactions + keyframes hit 84% and 99% — leto__jean · 2026-09-17