Lorcan is a small quadrupedal robot designed to be relatively cheap to produce (< $1k USD) and agile.
It is inspired in part by this post. However, I wanted to make my own version that is quasi-direct drive, using a simple gear system.
Lorcan uses 4 ODrive motor controllers, and 8 encoders.
The intent is to use machine learning to learn to walk instead of hand coded control systems. It will use a bottom-facing camera with optical flow to generate a reward signal for reinforcement learning. I plan to use a Raspberry Pi 4.
A small issue in the design requires the printing of 4 additional spacers. These could have been baked into the femurs, but having them be separate works as well. I have uploaded these spacers. I have also improved the tolerances for the ball bearing slots in the extensions (I already have them with the worse tolerances, but I updated it in case anyone else would like to print it).
I have included a new progress shot. The robot mechanics are done, wiring and electronics remain.
I have completed the robot (see image), I can control the limbs arbitrarily. Still working on some basic walking gate to get things started, after which I will switch to using OgmaNeo for reinforcement learning (after mounting a Pi Camera).
This post will be updated as the project progresses.