An apparatus and method for selecting a plurality of terminal devices for uplink and downlink transmissions are described. A first terminal device is selected from a first group of terminal devices. A reinforcement learning agent is used for selecting the first terminal device based at least on data rates of the terminal devices of the first group and data rates of the terminal devices of a second group. The selected first terminal device is transferred from the first group of terminal devices to the second group of terminal devices. One group of terminal devices is scheduled for downlink transmission and the other group of terminal devices is scheduled for uplink transmission.