This repository has been superseded by L4acados, which contains an improved zero-order GP-MPC implementation.

---

Zero-Order GP-MPC

An efficient implementation of a tailored SQP method for learning-based model predictive control with ellipsoidal uncertainties,

• using the acados Python interface to solve the optimal control problems, and
• employing PyTorch to evaluate Gaussian process dynamics models with support for GPU acceleration,
• efficiently handling the optimization over the posterior covariances of the GP inside the optimizer,

by modifying the Jacobians inside the SQP loop directly.

Background

This software is built upon the results of the article "Zero-Order Optimization for Gaussian Process-based Model Predictive Control", published in the ECC 2023 Special Issue of the European Journal of Control (EJC), available at https://www.sciencedirect.com/science/article/pii/S0947358023000912. The code for the numerical experiments in the publication can be found here.

Installation instructions

1. Clone this repository.

       git clone https://github.com/lahramon/zero-order-gpmpc.git

2. Initialize submodules.

       git submodule update --recursive --init

3. Build the submodule acados according to the installation instructions.

       mkdir -p acados/build
       cd acados/build
       cmake -DACADOS_PYTHON=ON .. # do not forget the ".."
       make install -j4

4. Set up Python environment (Python version 3.9.13). In the following we use pipenv.

   1. Change into main directory, then run

          pipenv install

   2. Export the variables ACADOS_SOURCE_DIR and LD_LIBRARY_PATH point towards the right locations in your acados installation:

          export ACADOS_SOURCE_DIR=<yourpath>/zero-order-gpmpc/acados
          export LD_LIBRARY_PATH=<yourpath>/zero-order-gpmpc/acados/lib

      With pipenv, this can be done by defining a file called .env in the root directory, e.g., <yourpath>/zero-order-gpmpc/.env, which contains

          ACADOS_SOURCE_DIR=<yourpath>/zero-order-gpmpc/acados
          LD_LIBRARY_PATH=<yourpath>/zero-order-gpmpc/acados/lib

Examples

You can find an example implementation of the zero-order GP-MPC method here

Citing us

If you find this software useful, please consider citing our corresponding article as written below.

@article{lahr_zero-order_2023,
  title = {Zero-Order optimization for {{Gaussian}} process-based model predictive control},
  author = {Lahr, Amon and Zanelli, Andrea and Carron, Andrea and Zeilinger, Melanie N.},
  year = {2023},
  journal = {European Journal of Control},
  pages = {100862},
  issn = {0947-3580},
  doi = {10.1016/j.ejcon.2023.100862}
}