Design of Safe Optimal Guidance With Obstacle Avoidance Using Control Barrier Function-Based Actor–Critic Reinforcement Learning
针对无人机和导弹等飞行器在复杂环境中拦截机动目标时的避障问题，提出一种结合控制障碍函数和演员评论家强化学习的安全制导方案，通过扩展扰动观测器在线估计目标信息，并利用高阶控制障碍函数保证避障安全。
In path planning and guidance algorithms for vehicles, such as unmanned aerial vehicles (UAVs) and missiles, it is essential and imperative to account for obstacle avoidance in complicated flight environment (e.g., no-fly zones). This article presents a novel safe guidance scheme with the guarantee of critical obstacle avoidance for intercepting maneuvering targets. First, the engagement of the missile and the target is formulated as nonlinear planer pursuit-evasion dynamics, and the interaction between the obstacle and the missile are determined. Then, the maneuvering target knowledge is estimated online by the extended disturbance observer (EDO) and incorporated into a separated feedback term of the input channel for the implementation of an approximately optimal actor–critic guidance law framework. To render the safety, the obstacle is thought as constraint objects and mathematically described by high-order control barrier functions (HO-CBFs). Furthermore, HO-CBFs are synthesized with the proposed guidance framework to intercept the maneuvering target with obstacle settings. Finally, numerical simulations under various types of nonstationary targets are performed to illustrate the feasibility and effectiveness of the proposed scheme.