Cart-Pole Reinforcement Learning

Balancing a typical inverted pendulum with Temporal Difference methods. Done as a personal hobby project. Controlled with a Raspberry Pi 1B using I2C interface. There is no x-position state sensing - the cart was smashing L/R end stops (and failing the episode), yet it eventually learns to throttle down such movements.

I used these ICs to control stepper:
- NXP PCA9629A
- TI DRV8825

Unfortunately the motor (pulled from my 3D printer) is not powerful enough for auto pole swing-up even with trajectory energy shaping methods.

Видео Cart-Pole Reinforcement Learning канала Cheng-Yueh Liu