This repository contains the source code for DirCol5i, a trajectory optimization library for Matlab. DirCol5i uses a medium-order direct collocation method that is designed to work well for problems that have high-order derivatives, such as jerk or snap, in either the objective or dynamics function. It also happens to automatically handle implicit second-order dynamics, as long as they are smooth.
I wrote this library in response to a question from Andy Ruina: Why do all trajectory optimization methods use explicit first-order dynamics?
- The short answer is that the explicit first-order form improves the sparsity pattern in the gradients, which in turn improves convergence properties of the method.
- As a result, this code works fine for standard trajectory optimization problems, but it is a bit on the slow side.
- That being said, it does seem to work better than standard algorithms on problems with high-derivatives and implicit dynamics.
For standard trajectory optimization in Matlab I suggest using either:
- OptimTraj - My general-purpose open-source trajectory optimization toolbox: https://github.com/MatthewPeterKelly/OptimTraj
- GPOPS-II - State-of-the-art professional trajectory optimization in Matlab: http://www.gpops2.com/
Let me know if you have any questions about this software, or if you are interested in using it for a research project.
If you use this code and would like to add some feature, or improve the code or the documentation, then make a pull request! If it is well done then I will be happy to accept it.
If you are using this software, please cite the accompanying paper: "DirCol5i: Trajectory Optimization for Problems With High-Order Derivatives", which can be found here: https://doi.org/10.1115/1.4041610
DirCol5i_TechnicalBrief_Sep19_2018.pdf
The documentation here is a bit sparse. If you are curious about this software and run into trouble, let me know and I'll add documentation as needed. Even better, if you are using this software and have some improvements to the documentation, then make a pull request!