Bob was created using matter.js for the physics and a very simple neural network, with no hidden layers or biases. I did not use any Machine Learning libraries for this; I did it from scratch in JavaScript (see ml.js). This uses very simple evolution: Bobs die if their head goes below the red line, and the ones that move furthest may reproduce and evolve. Also, the activation function is a sine wave because it works best that way. Surprisingly, after hours of intense coding and tweaking, Bob actually learned how to run and skip (I call it running and skipping, it's not exact)! This project is of the most complex ones I have ever done, and I am honestly shocked that I got it working. However, I can't stop it from falling after around 4 steps. This is from a competition between me and n8progrmas, where we both tried to make walking bots. Stay tuned for a video about this project, and enjoy!
If you want to see it in action, go to my website.
Note: the robot falls a few seconds after the video ends. :D