Unmanned driving decision and control method based on federal deep reinforcement learning

The invention discloses an unmanned driving decision and control model training method based on federal deep reinforcement learning. The method comprises four steps: initialization, data processing, deep reinforcement learning of unmanned driving decision and control of a client, and federal learning. According to the invention, federal learning training can be carried out on the premise of ensuring that client data are not out of the local, and the effect of making decisions and controlling the unmanned vehicle in different scenes is achieved. According to experimental tests, the unmanned vehicle can complete driving in different test scenes, and more stable speed and vehicle control can be maintained. 本发明公开了基于联邦深度强化学习的无人驾驶决策与控制模型训练方法，该方法一共分为四步：初始化、数据处理、客户端无人驾驶决策与控制的深度强化学习、联邦学习。本发明能够保证客户端数据不出本地的前提下，进行联邦学习训练，达到对无人车在不同场景下进行决策与控制的效果。实验测试，无人车能够在不同测试场景下完成驾驶，并且能够保持更稳定的速度及车辆控制。