Scope creep is the expansion of your project goals. When I started I had no education or experience in robotics. Some would argue I still don’t. But I had a few personal goals in 2016. One of those was to pursue my goal of learning about robotics.
It all stared with the realization that maybe I could automate many simple jobs I was paying $20 an hour for someone to perform. Like pushing a concrete saw. I have the gas powered version of this saw below.
At first I needed a lot of time to learn how to code, about motors, etc. I built a cnc machine from scratch so that I could warm up and gain some skill. Then when I was ready to proceed on the concrete saw project I realized that I could simply build a segway style balancing robot that would be able to push multiple walk behind tools. It only makes sense right? So then I built this thing. Its heavier than a buick. It worked alright but it needed arms.
When I realized how hard arms would be I thought to myself might as well just make legs to go with it. So I started building a robot that would be capable of doing basic tasks. Pushing things, stacking things etc. After designing a basic strong but lightweight framework, gearboxes and motor controller board I built an arm and a head to try out some basic tasks.
But how to code it? The first steps were obvious! After piping the data from the robot to my desktop computer and piping the motor controller data back out to the robot I once again I found myself in trouble. Because I had to dynamically program the robot to do everything. How could I teach my robot without explicitly coding for every situation it might encounter.
Ahah, machine learning and AI to the rescue. So now all I have to do is write a piece of software that pushes concepts into vector space, associates cnn filters and state transitions with that vector space, pushes all that into a latent space where I can use bagging, binning and gaussian process regression to get the robot to learn and act on voice commands. So I’m pretty sure I’m almost done…
I think the lessons learned was I should have just payed the guy the $20 an hour.