Modeling for Reinforcement Learning and Optimal Control: Double pendulum on a cart


Modeling is an integral part of engineering and probably any other domain. With the popularity of machine learning a new type of black box model in form of artificial neural networks is on the way of replacing in parts models of the traditional approaches. However we think that this does not mean that traditional models will be less significant, but they might get even more important in some domains. Traditional models can for example be used to feed the deep learning algorithms that (at the moment) are hungry for large amounts of data, by generating data using classical modeling approaches.

Release of OpenOCL v4.20


With the release of OpenOCL 4.20 the non-linear program which is constructed from the optimal control problem definition now has a block sparse structure. This should enable the use of better structure exploiting solvers. Ipopt’s performance might improve as well. Parameters will not destroy the block sparsity structure.

Release of OpenOCL v4.00


We release OpenOCL 4.00 which from now on only support one way to define a dynamical system and optimal control problem. The old style of inheriting from OclSystem and OclOCP is not supported anymore. Get the new version of the toolbox here.

[Contributed] 1:10 scale pro stock remote car racing


Mustafa Alp from the Polytechnic University of Milan sent us a great video of his implementation with the OpenOCL toolbox.

Minor release OpenOCL v3.20


Minor release OpenOCL 3.20 with a fix for the Simulator. Download the toolbox from here. In the upcoming weeks we are expecting a major release where the backend of OpenOCL is implemented in C++, and following that OpenOCL will become a Python interface.

New release OpenOCL v3.11


We are happy to announce that OpenOCL v3 is ready and released. It contains many improvements with respect to the previous version and also some API changes. If you come from an earlier version see the release notes that contains a list of changes below. Otherwise you can directly go to the download page of OpenOCL v3.11 or pull the latest version from the master branch.

API docs are now on the website!

Go to api-docs to view all classes, methods, and functions that you need to implement optimal control problems with the OpenOCL toolbox. The API should be almost complete for all public functions. It will be gradually extended as the toolbox evolves.