Myriad is a real-world testbed that aims to bridge the gap between trajectory optimization and deep learning. Myriad offers many real-world relevant, continuous-time dynamical system environments, and several trajectory optimization algorithms. These are all written in JAX, and as such can be easily integrated into a deep learning workflow. The environments and tools in Myriad can be used for trajectory optimization, system identification, imitation learning, and reinforcement learning.

We hope Myriad will help foster increased interest within the machine learning community for the development and application of learning algorithms in impactful real-world settings.

For an overview of the motivation and capabilities of Myriad, see our arXiv preprint.

For implementation and usage details, see the Myriad documentation.

To run the examples and experiments in Myriad, you must first set up the environment. This involves installing ipopt on your system, as well as installing the Myriad dependencies.

apt install coinor-libipopt-dev

brew install pkg-config
brew install ipopt

python3 -m venv env
source env/bin/activate
python -m pip install -U pip
python -m pip install -r requirements.txt

Now that you have installed the necessary dependencies, you can run some examples using the tools in Myriad. The easiest place to start is the Jupyter notebooks we provide. To run these, you'll need to install Jupyter notebook as well as the Myriad package itself, which will enable the notebook to see the packages it needs.

We can install Jupyter notebooks and Myriad by running

pip install notebook
python setup.py build
python setup.py install

Then we can run the notebooks by navigating to the notebook directory and running launching a notebook

cd myriad/examples
jupyter notebook

To see what is going on "under the hood" in the various examples mentioned above, you can look, modify, and run the code in the experiments directory.

For example, to run trajectory optimization, uncomment line 54 of run.py, and then run

python run.py

You can modify hyperparameters either by changing the defaults directly in the config.py file, or by passing them in as arguments to the run.py script. For example, you can try the following:

python run.py --system=SIMPLECASE --optimizer=SHOOTING
python run.py --system=CARTPOLE --optimizer=COLLOCATION --intervals=100
python run.py --system=VANDERPOL --optimizer=SHOOTING --intervals=1 --controls_per_interval=50
python run.py --system=CANCERTREATMENT --optimizer=SHOOTING --max_iter=500

To run the other experiments, such as system identification (with parametric or Neural ODE model) or end-to-end control-oriented imitation learning (with parametric or Neural ODE model), try uncommenting the corresponding line in run.py (60, 63, 69, 72), and looking at the corresponding code in the experiments directory.

├── myriad
    ├── examples        # Various examples presented in Jupyter notebooks
        └── ...
    ├── experiments     # Various experiments (see Section 6 of arXiv paper)
        └── ...
    ├── gin-configs     # Default environment parameters
        └── ...
    ├── neural_ode      # Infrastructure for Neural ODE dynamics models
        └── ...
    ├── nlp_solvers     # Nonlinear Program Solvers
        └── ...
    ├── systems         # Environments
        └── ...
    ├── trajectory_optimizers     # Trajectory optimization routines
        └── ...
    ├── config.py       # Training and environment hyperparameters
    ├── custom-types    # Project types
    ├── defaults        # Learning rates and SysID parameter guesses
    ├── plotting.py     # Code for plotting results
    ├── study_scripts.py   # To study learned dynamics and effect of noise
    ├── useful_scripts.py  # Scripts for various tasks
    └── utils.py        # Simple helper functions
├── run.py              # Entry point for running code
├── setup.py            # To install Myriad
└── tests               # Tests
    └── ...

• Lenhart et Workman, Optimal Control Applied to Biological Models. Chapman and Hall/CRC, 2007.
• Betts, Practical Methods for Optimal Control and Estimation Using Nonlinear Pcrogramming. SIAM Advances in Design and Control, 2010.
• Kelly, An Introduction to Trajectory Optimization: How to Do Your Own Direct Collocation. SIAM Rev., 2017.

@article{howe2022myriad,
  title={Myriad: a real-world testbed to bridge trajectory optimization and deep learning},
  author={Howe, Nikolaus and Dufort-Labb{\'e}, Simon and Rajkumar, Nitarshan and Bacon, Pierre-Luc},
  journal={Advances in Neural Information Processing Systems},
  volume={35},
  pages={29801--29815},
  year={2022}
}