About this Blog

I come to the field of robotics with a somewhat unique background. In the years it took for me to graduate with my bachelor’s degrees from Humboldt State University, I supported myself by working in the woods as a logger. All of those years working as a rigging man on cable logging operations has left me with a deep appreciation of the complexity of the tasks that individuals engaged in skilled manual labor must perform. Also working with, around and on heavy equipment has given me an appreciation for what it takes to run machinery as well.

Very few if any of those individuals that are working at the academic level of robotics engineering have ever spent any time in their life actually doing, for a living, the jobs that they are attempting to reproduce robotically. As a result, their notions about how easily certain jobs will be replaced with robots are drawn from only the most trivial levels of knowledge of what those tasks actually entail.

Because of this two things happen. First, because their understanding of the engineering problem is based on incomplete or even a false knowledge, their efforts will fail to produce a working solution. And second, publically expressing their opinions on how easily certain occupations will be replaced with robots only comes across as condescending and patronizing to those working people whose careers are on the receiving end of such engineering efforts.

The truth is, there is no robotic system in existence that can perform with the speed and dexterity of a human being engaged in the hand-picking of a food crop. And given current technology, it’s not even on the horizon that such a robot could be built.

While, to an outside observer, what a field worker does may appear as mindlessly repetitive, it is in fact anything but. Every plant is different. Every row is different. Even the same field on different days, is different. Once a robot leaves the factory floor and moves out into the farm field, the mine shaft, the logging site, or the construction site, everything changes.

A factory floor mounted robot can be programmed with a fixed tool-path, since everything the robot needs to do can be placed with the necessary precision ahead of time. But once you move outdoors, the ability to control the placement of tasks no longer is an option; so any robot that is intended to be run in an industrial situation outdoors has to be able to continually adjust its programming on a moment by moment basis. In other words, any agricultural robot, intended to replace the field worker picking food crops, will have to be able to run in a continuously learning mode.

This is where my interest in agricultural robotics turns into an interest in machine learning. And then by extension, to the question of what underlying computational hardware architectures lend themselves to machine learning. So, much of my interest in agricultural robotics overlaps with an interest in machine learning and hardware design.

While these will be the main topics of discussion on my blog, my interest doesn’t stop there. As a physicist, I’m deeply excited about the Theory of Computation, so I will, from time to time, post on more abstract elements of machine learning as well.

Glen Reuschling.