Researchers have developed an agent-centered architecture for general robot learning called Teach-and-Grow Learning (TGL). This approach allows robots to learn from a few successful demonstrations and create reusable 'Skill Blocks' that can be composed in new scenes. The TGL architecture includes a Skill Library to store executable behaviors and Experience Memory to carry forward successes, failures, and repairs. This method enables robots to acquire new tasks without retraining policies, reducing the need for extensive data collection and policy updates. --- Why it matters: This matters because it could significantly reduce the 'retraining tax' in robotics, which is the burden of collecting new data and updating policies when a robot encounters an unfamiliar object or situation. By allowing robots to learn from experience and adapt to new tasks, TGL has the potential to improve the efficiency and effectiveness of robotic systems. Source: https://arxiv.org/abs/2608.17209