Robotics Social Media and Updates

Language‑Based Policies Enable Zero‑Shot Cross‑Robot Skills

Same task, same scene. Different robot. Everything breaks… LAP trains the model to express motion in plain language, for example “move forward 3 cm” or “tilt 15 degrees” instead of predicting joint torques or tokens. Because language carries structure and meaning, the policy learns behavior rather than memorizing one robot’s mechanics. That makes the same model usable across different embodiments without retraining. Results on real robots: About 50% zero-shot success on unseen platforms, around 2× stronger than prior VLAs, and faster adaptation when fine-tuned. In practice this points to a shift in robotics: scaling data helps, but shared representations matter more. If robots speak the same action language, they can share skills. Thanks for sharing, @LihanZha Project https://lap-vla.github.io/ Code https://lnkd.in/dQ6EkwQ5 —— Weekly robotics and AI insights. Subscribe free: 22astronauts.com