LatentAM: Real-Time Large-Scale Robotic Semantic Mapping via Online Dictionary Learning
rsasaki0109 · x · 2026-08-15
LatentAM is an online 3D Gaussian Splatting (3DGS) mapping framework for open-vocabulary robotic perception. It builds scalable latent feature maps via online dictionary learning, being model-agnostic and pretraining-free. Each Gaussian primitive has a compact query vector converted to approximate VLM embeddings via attention with a learnable dictionary. Voxel hashing manages maps, with GPU optimizing active local map and CPU storing global map. Experiments show superior feature reconstruction fidelity and near-real-time speed (12-35 FPS).
More from Embodied
- Robotics Acceptance Challenge: Defining "Task Complete" — CyberRobooo · 2026-08-15
- dots3 preview: Long-horizon agency in real life — otarU · 2026-08-15
- Ryan Shrout Shows Off New NVIDIA GeForce Hardware — ryanshrout · 2026-08-15
- Apple Hires Researcher to Bring Advanced Video Reasoning Models to iPhone — CSProfKGD · 2026-08-15
- reBot Arm Control Stack Integrates Agentic AI, VLM, and LLM for Robotics — kamathsblog · 2026-08-15
- Indians paid ₹250/hour to film chores at home, feeding training data for humanoid robots — markjeffrey · 2026-08-15