Agent behavior modeling method based on reinforcement learning and human in the loop

Computer generated force (CGF) is one of the increasingly important research topics in field simulation. However, low modeling efficiency and lack adaptability are acute problems traditional CGF modeling. In this study, a method for agent behavior based on reinforcement learning human loop proposed to improve ability First, an overall framework intelligent agents constructed deep algorithm Soft Actor Critic (SAC) framework. Second, order overcome slow convergence speed SAC framework, interaction value evaluation introduced, specific flow designed. Third, verify performance method, experiments conducted compared with algorithms using pure example completing tasks. Result shows that after 100 episodes training, task completion rate can approach 100% while require at least 500 training gradually rate. Finally, results demonstrate significantly increases number interventions loop.