Bayesian Inference for Path Following Control of Port-Hamiltonian Systems with Training Trajectory Data

Abstract

This paper describes a procedure to design a path following controller of port-Hamiltonian systems based on a training trajectory dataset. The trajectories are generated by human operations, and the training data consist of several trajectories with variations. Hence, we regard the trajectory as a stochastic process model. Then we design a deterministic controller for path following control from the model. In order to obtain reasonable design parameters for a path following controller from the training data, Bayesian inference is adopted in this paper. By using Bayesian inference, we estimate a probability density function of the desired trajectory. Moreover, not only the mean value of the trajectory but also the covariance matrix is acquired. A potential function for path following control is obtained from the probability density function. By incorporating the covariance information into the control system design, it is possible to create a potential function that takes into account uncertainty at each position on the trajectory, and it is expected to construct a control system that generates appropriate assist force for a human operator.

Keywords

• path following control
• bayesian inference
• port-hamiltonian systems