Method and device for supporting maneuver planning for an automated driving vehicle or a robot

A method for assisting maneuver planning for transportation vehicle driving by automation or for a robot, the method comprising:

• determining optimal action values for discretized actions by dynamic programming based on a plurality of discrete states in a state space of an environment of the transportation vehicle or the robot;
• initializing a reinforcement learning agent based on the optimal action values determined by the dynamic programming by learning a mapping between the plurality of discrete states in the state space, as input values, and action values for actions in the state space, as output values, wherein the discrete states and corresponding action values for discretized actions determined by the dynamic programming are used to initialize the reinforcement learning agent;
• following initialization of the reinforcement learning agent, using the reinforcement learning agent to perform a reinforcement learning method to adjust the learned mapping built based on the action values for the discretized actions determined by the dynamic programming;
• and
• providing the adjusted learned mapping for the maneuver planning,
• wherein the state space is described in discrete form by a Markov decision process,
• wherein the providing of the mapping is provided by a neural network that is trained via supervised learning to initialize the reinforcement learning agent based on the determined optimal action values determined by the dynamic programming, and
• wherein, following the initialization of the reinforcement learning agent based on the determined optimal action values, action values are also provided subsequently for states of the state space lying between the plurality of discrete states.