Skip to Main Content
Learning is an essential capability which an intelligent agent can own. We exert the profit-sharing reinforcement learning method into the semi-autonomous agent system (SAS). The semi-autonomous agent has constrained autonomy, and we call it restriction property which makes the SAS more flexible and robust in the military simulation and ITS etc. Profit-sharing method is more robust and fit for the dynamic environment which includes much uncertain factors, especially in the partial MDPs (Markov decision processes) environment, such as the battlefield. We propose an improving reinforcement learning method of profit-sharing used in the SAS. The new algorithm has its roots in the trait of SAS and the merit of profit-sharing reinforcement learning method. At last a trial of combatant searching safe blindage is introduced to prove the effectiveness of such approving profit-sharing method in the SAS.