动态环境无人机导航的安全分层强化学习框架

商益铭,杜昌平,杨睿,方天睿,杜泽安,郑耀

Safe hierarchical reinforcement learning framework for dynamic UAV navigation

Yiming SHANG,Changping DU,Rui YANG,Tianrui FANG,Ze’an DU,Yao ZHENG

表 2 自适应奖励塑形下的权重分配

Tab.2 Weight distribution under adaptive reward shaping

奖励组件 Global Safety Approach Goal 1.0 1.0 1.0 Direction 1.8 0.8 2.2 Collision 1.0 1.0 1.0 Boundary 1.0 1.0 1.0 Static 0.5 1.8 1.2 Dynamic 0.6 2.2 1.0 Deviation 0.7 1.5 0.8 Speed 2.5 0.6 0.9 Energy 0.4 1.4 1.6 Time 0.6 1.2 1.6 DWA 0.6 2.8 1.8 CBF 0.8 2.0 1.6