AVRL (4Wheel Steering Autonomous Vehicle based on Reinforcement Learning)

This project is performed in Engineering Desing (MECH404)

Hardware : 4Wheel Steering Car Software : Autonomous Driving

1. 4Wheel Steering Car observes environment by 4 cameras.
2. PPO model(RL) decides action based on 4 distance to line from car.
3. 4Wheel Steering Car do action RL model decides.
   (Action mode is only composed by [Go Straight, Go Diagonal, Turn])

Hardware

Hardware include individual wheel controled by each teensy. RL model provides proper action to each teensy for one action.

Car used two Raspberry Pi to allocate tasks deciding actions & observing environments. (Two Raspberry Pi connect each other by Ethernet communication.)

Deciding action Raspberry Pi send action command to each teensy.

Software

Software consists of three parts.

1. vision process - 4 cameras get frames & distance from car to lines
2. simulation - 2D simulation to train RL autonomous vehicle model (openai gym)
3. sim2real - by using vision process & RL model, execute hardware properly

Straight Mode

Diagonal Mode

Zero Turn Mode