The article discusses the use of backpropagation to optimize Model Predictive Control (MPC) performance by solving a policy optimization problem. It introduces a method to handle losses of feasibility and provides convergence guarantees. The content covers differentiable optimization, conservative Jacobians, and the application of backpropagation in closed-loop trajectory optimization. The algorithmic procedures outlined ensure efficient computation of gradients for closed-loop optimization. Extensions include dealing with infeasibility scenarios and incorporating state-dependent elements in the MPC problem.
To Another Language
from source content
arxiv.org
Дополнительные вопросы