Intelligent fuel-optimal guidance strategy for small body flexible landing

In this paper, a design scheme for the small body lander using flexible structure is proposed, and an intelligent guidance strategy for the final landing phase is given according to the lander's specific dynamical characteristics. The flexible lander consists of several rigid units (nodes), on which the navigation equipment, actuators and other payloads are installed. These nodes are wrapped and connected by the flexible material to reduce the impact when the lander touches the surface of the small body, and prevent damage or accidents such as bouncing and overturning, improving safety of the landing mission. In order to study the dynamic features of the flexible lander, an equivalent model is established first, and the flexible connection between nodes is simulated through a spring-damper-torsion spring system. On this basis, a polynomial-based analytical guidance law (E-guidance) is applied to generate the fuel-optimal nominal landing trajectory for each node. Furthermore, to solve the problem of large motion and attitude error of nodes under the influence of flexible connection, the twin delay deep deterministic policy gradient algorithm (TD3) is used to generate a supplementary term of the nominal guidance strategy. The algorithm belongs to the deep reinforcement learning theory, it can perform deep network modelling for the complex dynamics of the flexible structure by learning the response features of the node's motion states to control commands. The numerical simulation result shows that the trained reinforcement learning agent can effectively reduce the motion error of the flexible lander, improving its attitude stability and landing accuracy.