Learning-Based Guidance for Pinpoint Lunar Landing

The technology of pinpoint landing ensures the accuracy of reaching the desired site, guarantees the safety of the spacecraft, and expands the detection area of interest. However, due to the thrust magnitude deviation of the lander engine, current guidance algorithms are difficult to consider optimality, real-time performance, and pinpoint accuracy simultaneously. To improve upon this, a learning-based explicit guidance algorithm is proposed. Firstly, powered explicit guidance (PEG) is employed to calculate the steering parameters, then the predicted downrange is fed back to calculate switch time of thrust with the dichotomy method. Focusing on the problem that the predicted downrange is larger than the actual one, a bias is added in the prediction to improve the landing accuracy. The steering parameters of PEG and switch time of thrust are stored as datasets to train a neural network for guidance. After training, these parameters can be updated onboard as soon as the actual thrust magnitude is identified. The proposed method only requires one neural network to obtain the steering parameters of PEG and switch time of thrust meanwhile, which is computationally efficient.