Walking problem of bipedal humanoid robot: comparison between model-based and learning-based method

It is a challenge to make a humanoid robot do perfectly bipedal walking, many obstacles will arise during the process. The problem of humanoid robots needing to be able to maintain the stability of their posture in order to continue walking movements in the presence of external disturbances like uneven ground or external force has been discussed in this paper. There are two different approaches to solving the problem: the model-based method and the learning-based method. In this paper, the non-linear state estimation framework and the PR-SRL algorithm are picked to represent the two approaches. By comparing the methodology, result, and analysis of the two methods, the conclusion that in some cases, the model-based method should be used, and in some other cases, the learning-based method is better can be drawn.