VLM-guided bipedal robot reads signs, dodges banana peels and picks paths like a human
ki_ki_ki1 · x · 2026-11-19
A demo thread by @kikiki1 shows a bipedal robot whose VLM marks hazards (exploding cans, banana peels, a person lying on the path) as no-step zones for the walking policy. It reads sign semantics—"Road work ahead" sends it onto grass, "Merge back" brings it back—and picks human-like routes around obstacles.
More from Embodied
- Robot thread: a new sign redraws its risk map while semantic-free baselines walk straight in — ki_ki_ki1 · 2026-11-19
- Inside the 'most famous kitchen' in robotics: a data-collection lab full of robot arms — DominiqueCAPaul · 2026-11-19
- Drivr: a voice-first Grok-powered cabin bot that turns your Tesla into an agent terminal — bfrench · 2026-11-19
- Musk says Tesla plans a family Model Y with no steering wheel or pedals — Polymarket · 2026-11-19
- Dev builds "iPhone Duo" device for managing AI agents, claims desktop-level productivity — zack_overflow · 2026-11-19
- AI agent autonomously performs brain surgery on a mouse, choosing therapy and target region — HealthcareAIGuy · 2026-11-19