VAMOS hierarchical VLA lets a robot self-navigate 2.35 km through Zurich
KyleMorgenstein · x · 2026-11-05
A team released WayTrace, giving a robot only web-map-style walking directions and letting it autonomously navigate real Zurich streets — crosswalks, stairs, crowded streets, construction sites — covering 2.35 km total.
The core tech comes from the paper VAMOS (arXiv:2510.20818), a hierarchical vision-language-action model that decouples semantic planning from embodiment grounding:
- A generalist planner trained on diverse open-world data proposes candidate paths directly in image space;
- A specialist affordance model, trained in low-cost simulation, learns the robot's physical constraints and re-ranks those paths;
- This enables cross-embodiment navigation (legged and wheeled robots) and easy steering via natural language.
Real-world experiments show higher success rates than SOTA model-based and end-to-end methods in both indoor and complex outdoor navigation. VAMOS co-author Mateo Guaman Castro noted kilometer-level city navigation is extremely hard due to countless novel scenarios, and praised the team for turning the original ideas into reality.
More from Embodied
- OpenClaw robotic arm works as release manager, cherry-picking thousands of PRs — vincent_koc · 2026-11-05
- PyRUA-REAL: coding agents give robots recursive self-improvement in the real world — Qiang Wang · 2026-11-05
- Europe's robot-dense industry eyes grid inspection, but regulation stalls deployment — ingliguori · 2026-11-05
- Bambu Lab launches TPU 75A: softest filament yet for 3D-printed wearables — vincent_koc · 2026-11-05
- NVIDIA launches Isaac Sim tutorial series: from first robot to HIL deployment — lukas_m_ziegler · 2026-11-05
- China blocks SoftBank's $5.4bn takeover of ABB's robotics division — davidyin44 · 2026-11-05