FaSTrack

FaSTrack (Fast and Safe Tracking): fast planning methods with slower, reachability-based safety guarantees for online safe trajectory planning. Auto-generated documentation may be found here. fastrack is primarily developed by David Fridovich-Keil, a PhD student in the Hybrid Systems Lab and Berkeley AI Research.

Repository organization

All code in this repository is written in the Robot Operating System (ROS) framework, and as such is broken up into atomic packages that implement specific functionality. The ros/ directory is the root workspace, and individual packages live inside the ros/src/ directory.

Usage

First, make sure you have ROS installed on your system. The project was developed in Jade, but it should be compatible with anything past Hydro. Please let us know if you have any compatibility issues.

The core fastrack, fastrack_msgs, and fastrack_srvs have no significant external dependencies other than those listed below. However, the fastrack_crazyflie_demos package depends upon the crazyflie_clean repository, which contains drivers and utilities for the HSL’s Crazyflie 2.0 testbed.

Other dependencies:

• Gtest – Google’s C++ unit testing library
• Eigen – a header-only linear algebra library for C++
• OMPL – an open C++ library for motion planning (recommend v1.2.1 to avoid g++5 dependency)
• MATIO – an open C library for MATLAB MAT file I/O
• FLANN – an open source library for fast (approximate) nearest neighbors

To build the entire workspace, you must begin by building and sourcing the crazyflie_clean repository. Instructions may be found in that project’s README. Once you have done that, open a terminal window and navigate to the ros/ directory. Then run:

catkin_make

If you only wish to build the fastrack core packages (not fastrack_crazyflie_demos) then you may instead run:

catkin_make --pkg=fastrack

Every time you open a new terminal, you’ll have to tell ROS how to find this package. Do this by running the following command from the ros/ directory:

source devel/setup.bash

To run unit tests, type:

catkin_make run_tests

Citing this work

If you find this repository or the ideas implemented within it useful for your own work, please do check out the following papers and cite the appropriate ones:

• Original FaSTrack paper, Herbert 2017
• Planning, Fast and Slow, Fridovich-Keil 2018

C++ reference materials

We attempt to adhere to the philosophy put forward in the Google C++ Style Guide. Our code is written for the reader, not the writer. We write comments liberally and use inheritance whenever it makes sense.

A few tips, tricks, and customs that you’ll find throughout our code:

• Lines of code are no longer than 80 characters.
• The names of all member variables of a class end with an underscore, e.g. foo_.
• When iterating through a vector, we name the index something like ii instead of just i. This makes it super easy to find and replace the iterator later.
• We use the const specifier whenever possible.
• We try to include optional guard statements with meaningful debug messages wherever possible. These may be toggled on/off with the ENABLE_DEBUG_MESSAGES cmake option.
• Whenever it makes sense, we write unit tests for self-contained functionality and integration tests for dependent functions and classes. These are stored in the test/ directory.

Why so many templates?

Many of the important classes in fastrack are heavily templated. This may seem like an ugly way to write generic classes, and in some cases it is; however the logic behind it is that templating on the relevant types removes the need to use so many explicit pointers. For example, if a Planner operates on the QuadrotorDecoupled6D dynamics then it can actually store those dynamics as a member variable rather than as a std::unique_ptr<Dynamics> or something like that.

Inheritance is still used to specify interfaces, and in some cases provide (overridable) base class functionality. In some cases, parts of interfaces are specified only implicitly through templating, e.g. a line S::Dimension() will not compile if template type S does not provide a static function Dimension().