DATA DRIVEN ROBOT CONTROL

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for data driven robot control. One of the methods includes maintaining robot empirical data; acquiring annotation data; training a reward model on the annotation data; generating task-specific training data for the particular task, including, for each experience in a second subset of experiences in the robot experience data: processing an observation in the experiences using a trained reward model to generate a reward prediction, and associating the reward prediction with the experiences; and training a policy neural network on the task-specific training data for the particular task, where the policy neural network is configured to receive a network input comprising the observations and to generate a policy output defining a control policy for the robot to perform the particular task. 用于数据驱动机器人控制的方法、系统和装置，包括编码在计算机存储介质上的计算机程序。方法中的一个包括：维护机器人经验数据；获取注释数据；在注释数据上训练奖励模型；为特定任务生成特定于任务的训练数据，包括，针对机器人经验数据中经验的第二子集中的每个经验：使用训练