Quasistatic Simulator C++

Model paths

Until this repo and quasistatic_simulator are merged, the relative paths between this repo and quasistatic_simulator need to be manually maintained so that the model files in quasistatic_simulator can be found by the C++ backend in this repo.

Specifically, it is assumed that

• the get_model_paths.cc file of this repo can be found at

${HOME}/ClionProjects/quasistatic_simulator_cpp/src

• The quasistatic_simualtor repo can be found at

${HOME}/PycharmProjects/quasistatic_simulator

• The robotics_utilities repo can be found at

${HOME}/PycharmProjects/robotics_utilities

Eigen 3.4

Eigen 3.4 supports useful syntax such as slicing, so we'd like to use it. However, drake by default uses the Eigen bundled with the OS. Mac's default Eigen is 3.4, so there is nothing we need to do. On ubuntu 20.04, however, the system's default Eigen is 3.3.*, which means we need to ask drake to use a different, user-installed Eigen.

To build drake with a use-installed Eigen, one needs to turn on WITH_USER_EIGEN when building drake, and set Eigen3_DIR to the folder containing cmake config files of the user-installed Eigen. If Eigen is installed using the default options with the .tar ball downloaded from Eigen's official website, Eigen3_DIR should be /usr/local/share/eigen3/cmake. In contrast, the system's Eigen can be found at /usr/lib/cmake/eigen3 instead.

To build this project with the user installed Eigen, add this flag when running cmake:

-DEigen3_DIR=/usr/local/share/eigen3/cmake

Running tests

At the root of this repo, run

mkdir build && cd build
cmake .. # with -DCMAKE_PREFIX_PATH=/path/to/drake and -DCMAKE_BUILD_TYPE=release, if necessary.
make test 

When built in release mode, the TestGradient* tests may fail due to a handful of gradients solved in single thread is different from the corresponding gradient solved in parallel. This might be caused by the different behaviors of Eigen's BdcSvd in debug and release modes: the TestGradient* tests never failed in debug mode based on my testings. Therefore, it should be fine (for practical purposes) as long as the failures consist of only a few gradient differences exceeding tolerance.