Car-racing

Repo for solving car-racing gymnasium environment through Reinforcement learning.

Question 2 : External library algorithm

Training

In order to start the training of a model presented in the report, you can use the command

python src/sb3/interface.py \   
    --model "ppo" \             # "ppo" | "ddp" | "sac"
    --timesteps 7500 \          # number of timesteps used for training
    --eval_freq 2500 \          # frequence at which the model is evaluated
    --eval_episode 20           # number of episode on which the model is evaluated

The three last parameters are optional. As for the model, the accepted values are "ppo", "ddpg" and "sac"

• Starting PPO

python src/sb3/interface.py --model "ppo"

• Starting DDPG

python src/sb3/interface.py --model "ddpg"

• Starting SAC

python src/sb3/interface.py --model "sac"

Evaluation

• Evaluating PPO

python src/interface.py --model "ppo" -sb

• Evaluating DDPG

python src/interface.py --model "ddpg" -sb

• Evaluating SAC

python src/interface.py --model "sac" -sb

Question 3 : Personal algorithm

Training

• Training DDPG

python src/algo/ddpg.py --config cfg_agent/cfg_ddpg.yaml

• Training PPO

python src/algo/ppo.py --config cfg_agent/cfg_ppo.yaml

• Training Beta PPO

python src/algo/beta_ppo.py --config cfg_agent/cfg_ppo.yaml

Evalulation

• Evaluating PPO

python src/interface.py --model "ppo" 

• Evaluating PPO with beta distribution

python src/interface.py --model "beta_ppo" 

• Evaluating DDPG

python src/interface.py --model "ddpg"