MIT's DAAAM Uses VLMs to Build Real-Time 3D Dynamic Scene Graphs
tom_doerr · x · 2026-08-11
MIT introduced DAAAM (Describe Anything, Anywhere, at Any Moment), a novel approach designed to give robots real-time, large-scale spatio-temporal memory.
Core Technology:
- The method integrates SAM segmentation, BotSort tracking, and VLM grounding.
- These inputs are fed into the Hydra system to construct 3D Dynamic Scene Graphs on the fly.
Key Contributions:
- Proposes a novel optimization-based frontend for extracting semantic descriptions from localized captioning models.
- Achieves hierarchical 4D scene graph construction with real-time performance.
- Delivers state-of-the-art results on NaVQA and SG3D benchmarks.
More from Embodied
- The Robot Taking Your Job Won't Be Impressive, Just Cheaper — VraserX · 2026-08-11
- 200ms Cloud Latency is Fatal for Robots: Why Edge Inference is Mandatory — vaibhavbetter · 2026-08-11
- Robotics VLA Paper 'OAT' Named RSS 2026 Outstanding Paper Finalist — Haoyu_Xiong_ · 2026-08-11
- Robotaxi Set to Disrupt Ride-Hailing and Car Ownership with Lower Costs — mitchdeg · 2026-08-11
- $500 Robotic Arm Enables Fully Automated 3D Printing — FinanceYF5 · 2026-08-11
- Alibaba's Qwen Open Platform Launches with AI Smart Glasses Skill Development — 千问APP · 2026-08-11