Sample-efficient Low-level Motion Planning for Robotic Manipulation Tasks via Zero-shot Transfer Learning
Researchers have developed a new framework, iCEM+TL, to improve the efficiency of low-level motion planning for robotic manipulation tasks. This approach combines the Sample-efficient Cross-Entropy Method (iCEM) with Transfer Learning (TL) to transfer parameters from simpler tasks to more complex ones. The framework also incorporates Reward Redesign (RR) through task decomposition for specific actions like stacking and shelf placement. Simulations showed up to a 23% improvement in success rates, and the method was successfully demonstrated on a real Franka Emika robot. AI
IMPACT Enhances robotic manipulation capabilities by improving planning efficiency and success rates in complex tasks.