Legged robot using differential evolution and perception
This is what a simple Evolutionary algorithm can achieve without the complexity of Neural Networks. The top world is the robot's imagination of the actions it considers performing. The bottom world is the real robot. This was my idea of utilizing physics simulation environments like PyMunk to represent the robot's perceived world in the robot's memory, and allowing the physics simulations to handle the prediction of motion, rather than have to manually program it. This could allow robots to adapt to environments with different gravity, viscosity and friction, without having to change the code. Just the simulation parameters need to change.
Related video: • Robot imagination helping with tackling va...
Background music is from: Cynthesize - New Age_Cynthesize - New Age FreeBackgroundMusic soundcloud