Bitcoin Lightning Forest (btclnfrst) is an experiment in accepting Bitcoin Lightning Network payments to plant trees in a virtual forest:
The blog post about this experiment is here.
After watching an episode of Black Mirror (MetalHead), I started OpenK9 as an experiment in building a robotic dog using low-cost manufacturing (e.g. 3D printing), and Reinforcement Learning (RL) to develop the control system.
I developed the mechanical system using Autodesk Fusion with parametric modelling, so I could optimize the geometry in a feedback loop with the RL system (in theory). The parts were lasercut and I used decent servos, but the system had really bad torsional rigidity from the use of three rod for the "spine". Here's a video of the final mechanics:
The RL system was a bit more interesting. I used Mujoco and OpenAI gym to build a learning system for the leg movement. I did achieve some success in learning a running gait as you can see here:
The source code for the Mujoco/OpenAI loop can be found here: https://github.com/bjnortier/max
Ultimately the complexity was too much for a side project, I think it would have to be a Master's level project to have some success. I was quite interesting to lean about Reinforcement Learning using Tensorflow, and working on the mechanical design.