Yesterday I got the neural network controller of Balance Bot working! It's the culmination of many years of learning ML and improving my robot building skills. Here are some lessons learned around real-world Reinforcement Learning:
Why ML and not a simple controller? I built a Proportional-Integral controller that works OK but the imperfections in the robot build (slop) make it very non linear and PI fails to recover from high speeds, external forces, and jitter in the frame.
Oct 15, 2019 • 14 tweets • 3 min read
I read a research paper last night that I just can't stop thinking about.
The idea was to write a differentiable rigid body physics simulator.
I'm just blown away. I'm trying to express how big of an epiphany this has been for me, but I just can't find the words.
Let's start with this: when modeling control systems, you express them in 1 of 3 forms:
