A method for reinforcement machine learning uses a reinforcement learning system that has an environment and an agent. The agent has a policy providing a mapping between states of the environment and actions. The method includes: determining a current state of the environment; determining, using the policy, a current policy output based on the current state; determining, using a knowledge function, a current knowledge function output based on the current state; determining an action based on the current policy output and the current knowledge function output; applying the action to the environment resulting in updating the current state and determining a reward; and updating the policy based on at least one of the current state and the reward.