Iterative Linearized Control: Stable Algorithms and Complexity Guarantees
[edit]
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:55185527, 2019.
Abstract
We examine popular gradientbased algorithms for nonlinear control in the light of the modern complexity analysis of firstorder optimization algorithms. The examination reveals that the complexity bounds can be clearly stated in terms of calls to a computational oracle related to dynamic programming and implementable by gradient backpropagation using machine learning software libraries such as PyTorch or TensorFlow. Finally, we propose a regularized GaussNewton algorithm enjoying worstcase complexity bounds and improved convergence behavior in practice. The software library based on PyTorch is publicly available.
Related Material


