← Getting Started

Overview

In this section, we will discuss:

• the personal background of the auther

• a serious problem in robotics

• a proposed solution (hint: it may be pykal)

• the structure of the “Tutorial”” and how best to work through it

Who am I?

My name is Nehal Mangat, and I am a (handsome) graduate student at the University of Nevada, Reno, where I am pursuing a masters in Mechanical Engineering and Applied Mathematics. I am also a robotics research assistant in the van Breugel lab and a former intern of the University of Florida’s Autonomous Vehicles Laboratory.

I have been involved in the world of academic robotics for several years now. I love my field, but it is a shame to say that there is something rotten in the state of academic robotics.

The Problem

Academia has witnessed an explosion of control and estimation algorithms. This should be a good thing; certainly we should not be worse off for having more ideas.

Unfortunately, academia values novelty, not utility; that is, researchers are incentivized to publish novel results at the expense of useability. With little reason to test results in reality, the hardware implementation of an algorithm is passed onto future researchers – who, pressed to publish novel results of their own, instead extend the algorithm, publish, and pass the buck along to future researchers, ad infinitum.

When the day comes that the algorithm is finally implemented onto hardware, and it does not work as expected (or at all), where do we begin troubleshooting? Is it a hardware issue, a software issue, or does the real issue lie in the third paper of this algorithm’s journey, where the substitution of a new numerical method lead to unforeseen consequences when faced with asynchronous sensor data?

This is a tragedy of efficiency, to say nothing of the sanity lost in troubleshooting such issues.

The Solution

The pykal package is a Python package designed to make the following transitions as painless as possible:

ROS and Gazebo are mature open-source robotics platforms that have already eased the burden of the last bridge; pykal aims to ease the burden of the first two. How?

pykal uses the discrete-time dynamical system as a compositional primitive in defining algorithms and control systems. Such representations are modular, easy to implement, and easy to debug, thus spanning the gap between “Theory” and “Python””.

Further, a modular framework induces a simple API between Python and ROS, which lets us abstract away from the CLI and interface with ROS from a Jupyter notebook via pykal. In addition to a quality of life improvement, the composition of dynamical systems has a direct correspondence with ROS node architecturem, making it possible to implement and debug such architecturs in Python before automagically translating them into ROS. Thus, the second bridge from “Python” to “ROS” is spanned.

Using the Tutorial

The “Tutorial” is split into four sections.

Overview

If you have managed to skip to the end of this one-page overview without reading anything, I recommend scrolling back to the beginning and reading the whole overview.

Theory to Python

If you are new to this package and want to try casting algorithms/control systems as dynamical systems, start here.

Python to ROS

If you are comfortable using pykal to cast algorithms/control systems as dynamical systems and want to begin simulating things in ROS, start here.

ROS to Gazebo

If you are comfortable simulating things in ROS and to put it in Gazebo, start here.

← Getting Started