Method for training machine learning model to realize control rules

The invention describes a method for training a machine learning model to implement a control rule, having: for each of a plurality of control operations, delivering input data specifying a respective control scenario to the machine learning model, wherein the machine learning model outputs output data specifying a probability distribution of trajectory parameter values in response to the conveyed input data, sampling trajectory parameter values from the probability distribution, determining a trajectory from the trajectory parameter values and evaluating the trajectory and adapting the machine learning model, the probability that the sampling determines the trajectory parameter value of the trajectory with the higher evaluation is improved relative to the probability that the sampling determines the trajectory parameter value of the trajectory with the lower evaluation. 本发明描述了一种用于训练机器学习模型以实现控制规则的方法，具有：针对多个控制运行中的每一个，向所述机器学习模型输送指定相应控制场景的输入数据，其中所述机器学习模型响应于所输送的输入数据输出指定轨迹参数值的概率分布的输出数据，从所述概率分布中采样轨迹参数值，根据所述轨迹参数值确定轨