CMU Introduces SuperMap: A Living Spatial Memory for Embodied AI
rsasaki0109 · x · 2026-07-30
Carnegie Mellon University (CMU) presented SuperMap, accepted to RSS 2026, proposing a "living spatial memory" system for embodied AI.
- System Features: A training-free spatio-temporal SLAM system that perceives the world, remembers its evolution, and supports reasoning and action.
- Architecture: Fuses high-frequency geometric SLAM with asynchronous open-vocabulary perception to build a persistent semantic world model.
- 4D Scene Graph: Produces a queryable map carrying spatial and temporal information for every object, enabling visual-language navigation and long-horizon reasoning on real robots.
More from Embodied
- IROS 2026 Origami Challenge Dataset Released: 4.76M Frames of Tactile Data — DJiafei · 2026-07-30
- 8 Chinese Humanoid Robot Startups Hit $2.8B Valuation: Three Core Logics Decoded — 创业邦 · 2026-07-30
- Drones Inspect Wind Turbines in 30 Mins with AI-Powered Defect Analysis — lukas_m_ziegler · 2026-07-30
- Rabbit R1 Becomes 'Really Good' After Integrating Hermes — SimonBalmain · 2026-07-30
- German Startup Uses Private Chef's Cooking Footage to Train Humanoid Robots — Wired AI · 2026-07-30
- Festo Unveils Four-Winged Bionic Robot That Flies Like an Insect — TansuYegen · 2026-07-30