Deep Reinforcement Learning for Pin-Point Autonomous Lunar Landing: Trajectory Recalculation for Obstacle Avoidance

This work aims to present a method to perform autonomous precision landing—pin-point landing—on a planetary environment and perform trajectory recalculation for fault recovery where necessary. In order to achieve this, we choose to implement a Deep Reinforcement Learning—DRL—algorithm, i.e. the Soft Actor-Critic—SAC—architecture. In particular, we select the lunar environment for our experiments, which we perform in a simulated environment, exploiting a real-physics simulator modeled by means of the Bullet/PyBullet physical engine. We show that the SAC algorithm can learn an effective policy for precision landing and trajectory recalculation if fault recovery is made necessary—e.g. for obstacle avoidance.