Method for controlling an agent

1 . A method for controlling an agent, comprising the following steps:

training a neural network using training data that contain, for a multiplicity of agents, examples of a behavior of the agents, an output of the neural network including a prediction of a behavior and being a function of network parameters that are trained in common for all training data, and being a function of a further parameter that is trained individually for each of the agents of the multiplicity of agents;

fitting of a probability distribution to values of the further parameter for the agents that result from the training;

sampling a value from a probability distribution for a further agent in an environment of the agent; and

controlling the agent, taking into account a prediction of a behavior of the further agent that the neural network outputs for the sampled value for the further agent.

2 . The method as recited in claim 1 , wherein the neural network is trained to map an input of the neural network onto a prediction of a behavior of an agent, the input of the neural network including state information of the agent for which the neural network is to predict the behavior, and including the further parameter.

3 . The method as recited in claim 2 , wherein the input of the neural network includes state information about a control scenario in which the behavior of the agent is to be predicted.

4 . The method as recited in claim 1 , wherein the probability distribution is a Gaussian mixture model.

5 . The method as recited in claim 1 , wherein the training of the network parameters and of the further parameter is by adapting the network parameters and the further parameter in to minimize a loss between the examples of the behavior of the agents and the behavior respectively predicted by the neural network.

6 . A control device configured to control an agent, the control device configured to:

train a neural network using training data that contain, for a multiplicity of agents, examples of a behavior of the agents, an output of the neural network including a prediction of a behavior and being a function of network parameters that are trained in common for all training data, and being a function of a further parameter that is trained individually for each of the agents of the multiplicity of agents;

fit of a probability distribution to values of the further parameter for the agents that result from the training;

sample a value from a probability distribution for a further agent in an environment of the agent; and

control the agent, taking into account a prediction of a behavior of the further agent that the neural network outputs for the sampled value for the further agent.

7 . A non-transitory computer-readable medium on which are stored commands for controlling an agent, the commands, when executed by a processor, causing the processor to perform the following steps:

training a neural network using training data that contain, for a multiplicity of agents, examples of a behavior of the agents, an output of the neural network including a prediction of a behavior and being a function of network parameters that are trained in common for all training data, and being a function of a further parameter that is trained individually for each of the agents of the multiplicity of agents;

fitting of a probability distribution to values of the further parameter for the agents that result from the training;

sampling a value from a probability distribution for a further agent in an environment of the agent; and

controlling the agent, taking into account a prediction of a behavior of the further agent that the neural network outputs for the sampled value for the further agent.