An Improved Proximal Policy Optimization Method for Low-Level Control of a Quadrotor

In this paper, a novel deep reinforcement learning algorithm based on Proximal Policy Optimization (PPO) is proposed to achieve the fixed point flight control of quadrotor. The attitude and position information quadrotor directly mapped PWM signals four rotors through neural network control. To constrain size policy updates, PPO Monte Carlo approximations optimal penalty coefficient. A optimization method with penalized probability distance can provide diversity by performing each update. new proxy objective function introduced into actor–critic network, which solves problem falling local optimization. Moreover, compound reward presented accelerate gradient along update direction analyzing various states that may encounter in flight, improves efficiency network. simulation tests generalization ability offline changing wing length payload Compared method, has higher better robustness.