txia-m/ppo-MountainCar-v0-a01

PPO Agent playing MountainCar-v0

This is a trained model of a PPO agent playing MountainCar-v0 using the stable-baselines3 library.

Usage (with Stable-baselines3)

import gymnasium as gym
import torch
import numpy as np
from stable_baselines3 import PPO
from stable_baselines3.common.env_util import make_vec_env
from stable_baselines3.common.vec_env import VecNormalize


class StrictPhysicsWrapper(gym.Wrapper):
    def __init__(self, env):
        super().__init__(env)

    def step(self, action):
        next_state, reward, terminated, truncated, info = self.env.step(action)
        position = next_state[0]
        velocity = next_state[1]

        dist_from_bottom = abs(position - (-0.5))
        potential_reward = dist_from_bottom * 0.9

        kinetic_reward = (velocity ** 2) * 10.0

        new_reward = reward + potential_reward + kinetic_reward

        if position >= 0.5:
            new_reward += 100.0

        return next_state, new_reward, terminated, truncated, info


env = make_vec_env("MountainCar-v0", n_envs=16, wrapper_class=StrictPhysicsWrapper)
env = VecNormalize(env, norm_obs=True, norm_reward=False, clip_obs=10.)

model = PPO(
    policy='MlpPolicy',
    env=env,
    learning_rate=3e-4,
    n_steps=1024,
    batch_size=64,
    n_epochs=10,
    gamma=0.99,
    gae_lambda=0.95,
    ent_coef=0.0,
    verbose=1,
    policy_kwargs=dict(
        net_arch=dict(pi=[64, 64], vf=[64, 64]),
        activation_fn=torch.nn.Tanh
    )
)

model.learn(total_timesteps=1000000)

...