MAS-DDPG: Multi-Sensor Self-Adaptive Scheduling for UAV Path Planning in Network-Unavailable Environments
Na Lin, Zhijiang Wang, Liang Zhao · 2023
In network-unavailable environments, the path planning of unmanned aerial vehicles (UAVs) becomes constrained by limitations in their energy resources and computing capabilities within a predefined mission timeframe. Additional low-power flight control must be considered to effectively complete a flight mission. To support UAV path planning in network-unavailable environments, this paper proposes a multi-sensor self-adaptive scheduling and optimized depth deterministic strategy gradient (MAS-DDPG) algorithm. Specifically, unlike considering activating all sensors in varying scenarios, a multi-sensor self-adaptive scheduling (MAS) strategy based on Multi-nominal Logit (MNL) is proposed. This approach allows the UAV to choose an appropriate scheduling strategy according to different scenarios. Furthermore, instead of relying on randomly chosen experiences from a pool, a mechanism based on screening experience is proposed. This mechanism takes into account successful experiences along with their contextual counterparts to enhance the learning process. Our goal is to maximize the average flight speed of the UAV and minimize the number of flight decisions and flight energy consumption. Simulation results show that the proposed approach can achieve significantly hinter speed and lower energy consumption than the benchmark and other typical algorithms.