Artificial Neural Network Control of a Bipedal Robot Using a Bond Graph Model

The field of robotics is rapidly advancing, with particular focus on the study of bipedal locomotion, pushing the limits of what robots can achieve in terms of mobility and autonomy. Consequently, robotic locomotion has become a critical area of research in the development of bipedal robots. Various approaches exist for controlling bipedal gait cycles, utilizing different control techniques and advanced machine learning algorithms. Among these, artificial neural networks stand out as a highly effective supervised machine learning tool, used for prediction, classification, and control. In this paper, a successful gait cycle is defined by tracking the bipedal robot's center of gravity angle. An artificial neural network is employed as a controller to adapt the bond-graph model of the bipedal robot. This paper demonstrates the application of an artificial neural network as a machine learning algorithm to control and stabilize the locomotion of a bipedal robot with uniform response with overshot 0.25 for the linear velocity to overcome inertia and perform successful gait cycles.