机器人“干中学”，人类不用再给工厂中的机器人当保姆

Group 1 - The core viewpoint of the article highlights the successful implementation of real machine reinforcement learning technology by Zhiyuan Robotics in collaboration with Longqi Technology, which enhances the efficiency of robotic deployment on production lines [1][3] - Traditional reinforcement learning typically occurs in simulated environments, leading to challenges in transferring learned strategies to real machines, which often requires extensive adjustments and resources [1][2] - The deployment of humanoid robots in actual production lines is currently labor-intensive, with a significant number of personnel required for tuning, calibration, and safety monitoring [2] Group 2 - Directly embedding reinforcement learning into real production lines optimizes training objectives for robots, potentially reducing the need for human and material resources [3] - Despite the efficiency gains, there are risks associated with material loss and safety during the deployment of real machine reinforcement learning, necessitating pre-training and robust control mechanisms [3] - The next challenge for Zhiyuan Robotics is to replicate the success of real machine reinforcement learning across multiple production processes, leveraging local private cloud and OTA mechanisms for sharing learning experiences and model updates [3]