Optimal Dynamic Regret in LQR Control

Neural Information Processing Systems 

We consider the problem of nonstochastic control with a sequence of quadratic losses, i.e., LQR control.