residual model reinforcement learning

python main.py --log-dir log_ppo_base10 --mode train

python main.py --log-dir log_ppo_base12 --mode eval --action-bias 0.1

python main.py --log-dir log_ppo_base3 --mode residual_train --action-bias 0.5

--action-bias
--residual

1. train base policy python main.py --log-dir log_ppo_base15 --mode train --total-timesteps 6_000_000

2. eval in bias python main.py --log-dir log_ppo_base12 --mode eval --action-bias 0.5

3. learn residual in bias python main.py --log-dir log_ppo_base12 --mode residual_train --action-bias 0.1

4. train policy in residual python main.py --log-dir log_ppo_res_12 --mode train --residual

5. eval in bias python main.py --log-dir log_ppo_res_10 --mode residual_train --action-bias 0.5