Skip to Main Content
This paper focuses on the flight path planning process with multi-agent for Unmanned Aerial Vehicle (UAV) based on threats analysis and path length constraint. Path planner agent searches the path with global view considering path length constraint and information collector agent deals with path planning in the zone of threats. Scoring function is presented based on analysis the threats' attributes. We consider the path planning process as the multi-agent cooperation in a dynamic and non-stationarity environment. In order to perfectly adapt agents to environment changing, we restructure the traditional Q-value learning algorithm into a dynamic reinforcement learning algorithm by introducing current beliefs and recency based exploration bonus. The simulation results show that the proposed method converges rapidly and can be used in flight path planning.