OpenFrontier (RSS 2026): zero-shot open-world robot navigation with VLM-scored frontiers
rsasaki0109 · x · 2026-09-09
ETH Zurich's OpenFrontier (authors include Marc Pollefeys) is accepted at RSS 2026: "General Navigation with Visual-Language Grounded Frontiers," with code open-sourced on GitHub.
Key points:
- Goal: open-world navigation where robots pursue flexible goals in complex environments without task-specific training
- Method: zero-shot via visual frontiers as semantic anchors — sparse navigation targets detected in RGB, scored by an off-the-shelf VLM against the language goal, and maintained in 3D for global planning
- Codebase: modules for mapping, monodepth, navigation, planning, VLM/VLFM scoring, plus Habitat benchmark and demo scripts
A representative open-source work on VLM-grounded robot navigation.
More from Embodied
- LLMs With Mechanical Reasoning Plus CNC and Additive Manufacturing — granawkins · 2026-09-09
- Robot arm self-calibrates with 3 uncalibrated cameras, hits sub-0.2mm accuracy — burny_tech · 2026-09-09
- MicroPort installs Toumai surgical robot at Nigeria's Nisa Premier Hospital, first in Abuja — PAstynome · 2026-09-09
- Apple Quietly Acquired Sonera, Whose Chips Read Brain Activity via Magnetic Fields — ChuckDBrooks · 2026-09-09
- Jim Fan declares VLA dead: multimodal coding is the right action space for robotics' System 2 — DrJimFan · 2026-09-09
- Maker scales up the open-source Microduck robot into a bigger, clumsier Macroduck — kamathsblog · 2026-09-09