RADIO-ViPE enables open-vocabulary semantic SLAM from monocular video

Researchers have developed RADIO-ViPE, a novel semantic SLAM system capable of open-vocabulary grounding in dynamic environments using only monocular RGB video. This system integrates multi-modal embeddings from foundation models with geometric scene information, eliminating the need for depth sensors or pose initialization. RADIO-ViPE demonstrates state-of-the-art performance on the TUM-RGBD benchmark, offering robust semantic grounding for robotics and unconstrained video streams. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enables open-vocabulary semantic grounding in dynamic environments using only monocular video, advancing robotics and video analysis.

RANK_REASON Academic paper introducing a new system for semantic SLAM.

Read on arXiv cs.CV →

• RADIO-ViPE
• TUM-RGBD
• arXiv

• paper
• other