ENCORE Enables Few-Shot Discovery of Robot Manipulation Strategies
September 29, 2026
ENCORE allows robotic agents to learn manipulation tasks from a few demonstrations by distilling them into multi-view keyframes and gripper events. A coding agent then writes and refines a policy program against a fixed API to find successful grasping and contact strategies.
HOW THIS AFFECTS YOU
●
builderYou can implement robot task specification through high-level language using agentic policy generation.
●
researcherYou can explore few-shot agentic discovery as an alternative to traditional imitation learning.