Model Predictive Path Integral Control based on Model Sampling

In this paper, we describe a general procedure to use Model Predictive Path Integral(MPPI) control algorithm with model-sampling to solve classic robot tasks. First, the Gaussian Mixture Model(GMM) and the Linear Gaussian Model are used to build up the model from the sampled trajectories. Then, the MPPI algorithm utilizes the predicted information from the estimated model to generate a new trajectory with a high expectation of reward. We also tested the performance of the iterative Linear Quadratic Regulator(iLQR) algorithm and the MPPI with the real model as the control groups.