Developer Guide

I. About

Crocoddyl has been written in C++ for efficiency and uses the Eigen library for linear algebra routines. It comes with Python bindings for easy prototyping. Crocoddyl is currently supported for most Linux distributions, with plans to release on Mac OS X and Windows. The project is fully open-source under the permissive BSD-3-Clause license and is hosted on GitHub.

II. Design pattern

Crocoddyl is designed using virtualization pattern for easy prototyping and yet efficient implementation. In early benchmark, we have shown that virtualization is as efficient as static polymorphism (i.e. CRTP design) for system dynamics higher than 16. The real benefits of static polymorphism are in very small system (with dimension less than 6), however virtualization is still competitive.

One of the main design concepts of Crocoddyl is the strict separation between model and data. A model describes a system or procedure, e.g. action, cost, and activation models. Any model has been abstracted to easily derive new systems in both C++ and Python. By data, we refer to a container that stores computed and intermediate values used during the calculation routine performed by a model. Each model creates its own data, and with this, we avoid any temporary memory allocation produced by algebraic expressions in Eigen. To improve efficiency, especially in systems with dimension slower than 16, the Eigen members inside a data object can be defined with fixed dimensions. It allows for efficient use of modern CPU features by using vectorization and Same Instruction Multiple Data (SIMD) operations.

III. Code Format

Crocoddyl CI also checks code format (for both c++ and Python code) as effort to maximize readability and maintainability. Depending on the language, c++ or Python, we use different format policies which are tested with Gepetto/linters repository. For c++ code, Crocoddyl uses Google format convention, instead for Python code it follows flake8 rules with a maximum length of 119 characters.

Despite that Crocoddyl follows the strict code format, you don't need to learn each specific details of these formatting rules. For that, you have two options:

1. configuration your IDE to lint your code automatically (or check it), and
2. let our linter to adjust your code.

Here we explain how to do the later. For this, you can either use the provided docker image (subsection 1), or manually setup your environment by yourself (subsection 2).

1. Direct use of the provided docker image

docker run --rm -v $PWD:/root/src -it gepetto/linters --clang-6

This will not work if the root user doesn't have the right to go into your current working directory. And if your current working directory is accessed through symlinks, you might need to replace $PWD by $(pwd -P).

2. Manual use of each tool

Before starting with it, you need to clone Gepetto/linter repository, i.e.

git clone [email protected]:Gepetto/linters.git
sudo apt install clang-format-6.0
pip install --user flake8 isort yapf

Please note yapf is in a beta version, we tested our system with v0.28.0.

Please note that with the following instructions your current code will be updated

2.1. Formatting your c++ code

For c++ testing through clang-format. We used the configuration file defined in the Gepetto linter. Then, you need to install the configuration files in the current working directory (beware of git if doing so) or in a parent directory. Your home might be a good choice:

cd ${YOUR_CROCCODDYL_SOURCE_PATH}
ln -s ${YOUR_GEPETTO_LINTER_PATH}/.clang-format-6.0 .clang-format

And for formatting your c++ code, you just need to run the follows:

cd ${YOUR_CROCCODDYL_SOURCE_PATH}
clang-format-6.0 -i $(find . -path ./cmake -prune -o -iregex '.*\.\(h\|c\|hh\|cc\|hpp\|cpp\|hxx\|cxx\)$' -print)

2.2. Formatting your Python code

cd ${YOUR_CROCCODDYL_SOURCE_PATH}
flake8 .
yapf -ri .