Embodied Intelligence-Enhanced UAV Trajectory Design Via Deep Reinforcement Learning

As the low-altitude economy rapidly develops, unmanned aerial vehicles (UAVs) are emerging as key enablers of flexible communication services, particularly in environments that are only partially mapped and where users move dynamically. In such scenarios, trajectory design is highly challenging due to incomplete prior maps, uncertain obstacles, and time-varying user positions. To cope with these difficulties, we consider an embodied-intelligence (EI)-enhanced UAV that uses integrated sensing and communication signals to simultaneously serve moving users and sense its surroundings. By refining environmental knowledge from real-time echoes and relying on onboard decision intelligence, the UAV operates in a continuous sensing–decision–action loop, autonomously adjusting its flight path in response to environmental changes. Particularly, we formulate trajectory planning as a Markov decision process and develop an EI-enhanced deep deterministic policy gradient framework to determine the UAV's continuous actions under mobility and safety constraints. Simulation results demonstrate that the proposed scheme converges faster and yields better communication performance than baseline methods, validating its effectiveness in partially known environments with user mobility.