Safe learning-based gradient-free model predictive control based on cross-entropy method

In this paper, a safe and learning-based control framework for model predictive (MPC) is proposed to optimize nonlinear systems with non-differentiable objective function under uncertain environmental disturbances. The integrates MPC an auxiliary controller in way of minimal intervention. augments the prior nominal incremental Gaussian Processes learn cross-entropy method (CEM) utilized as sampling-based optimizer function. A intervention devised Lyapunov barrier guide sampling process endow system high probabilistic safety. algorithm shows adaptive performance on simulated quadrotor tasks trajectory tracking obstacle avoidance wind