Pure Perception Motion Control based on Stochastic Nonlinear Model Predictive Control

Noise coming from sensors or caused by external world phenomena results in measurement errors that cause uncertainties in some robotic tasks, e.g. tracking a robot displacement and tracking an observed target. Control approaches such as model predictive control (MPC) usually guarantee constraints satisfaction by way of using detailed models of prediction. Although the deterministic MPC allows certain robustness to be controlled in the system, it usually does not adequately deal with uncertainties. Therefore, we introduce in this manuscript a pure perception motion control, which consists of an approach that deals with the uncertainty problem through a stochastic nonlinear model predictive control (SNMPC) by minimizing only the covariances matrices of target observation and robot state estimation. As introduced previously, it can be used to track targets that are observed during some tasks. The SNMPC penalizes the undesired behavior, allowing the robot to converge to the optimal pose to observe the target optimally. A modification provided both in the prediction model and the cost function allows this minimization to be achieved. The proposed stochastic nonlinear controller is validated, providing a satisfactory control of the target tracking, by way of results obtained from simulation, which are presented and discussed in the paper to verify our proposal.