42 lines
888 B
Python
42 lines
888 B
Python
# %%
|
|
from stable_baselines3 import PPO
|
|
from stable_baselines3.common.env_util import make_vec_env
|
|
from intersim.envs.intersimple import ConstCollisionReward, IntersimpleFlatAgent
|
|
|
|
model_name = "ppo_const_collision"
|
|
|
|
class IntersimpleConstCollisionAgent(ConstCollisionReward, IntersimpleFlatAgent):
|
|
pass
|
|
|
|
env = IntersimpleConstCollisionAgent(
|
|
agent=51,
|
|
speed_reward_weight=0.001,
|
|
collision_penalty=1000
|
|
)
|
|
|
|
# %%
|
|
model = PPO(
|
|
"MlpPolicy", env,
|
|
learning_rate=3e-6,
|
|
verbose=1,
|
|
)
|
|
model.learn(total_timesteps=2e5)
|
|
model.save(model_name)
|
|
|
|
print('Done training.')
|
|
|
|
del model # remove to demonstrate saving and loading
|
|
|
|
# %%
|
|
model = PPO.load(model_name)
|
|
|
|
obs = env.reset()
|
|
while True:
|
|
action, _states = model.predict(obs)
|
|
obs, rewards, done, info = env.step(action)
|
|
env.render(mode='post')
|
|
if done:
|
|
break
|
|
|
|
env.close(filestr='render/'+model_name)
|