1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103import numpy as np
import rustoracerpy
import gymnasium as gym
from stable_baselines3 import PPO
from stable_baselines3.common.vec_env import VecEnv, VecMonitor, VecVideoRecorder
import wandb
from wandb.integration.sb3 import WandbCallback
from rustoracerpy.env import RustoracerEnv
import gc
class SB3VecAdapter(VecEnv):
"""Adapt a gymnasium VectorEnv to SB3's VecEnv interface."""
def __init__(self, venv: RustoracerEnv):
self.venv = venv
super().__init__(
num_envs=venv.num_envs,
observation_space=venv.single_observation_space,
action_space=venv.single_action_space,
)
self.metadata = venv.metadata
self._actions: np.ndarray | None = None
def reset(self) -> np.ndarray:
obs, _info = self.venv.reset()
return obs
def step_async(self, actions: np.ndarray) -> None:
self._actions = actions
def step_wait(self):
obs, rewards, terminated, truncated, infos = self.venv.step(self._actions)
dones = terminated | truncated
info_list = [{} for _ in range(self.num_envs)]
return obs, rewards, dones, info_list
def close(self) -> None:
self.venv.close()
def env_is_wrapped(self, wrapper_class, indices=None):
return [False] * self.num_envs
def env_method(self, method_name, *method_args, indices=None, **method_kwargs):
raise NotImplementedError
def get_attr(self, attr_name, indices=None):
return [getattr(self.venv, attr_name)] * self.num_envs
def set_attr(self, attr_name, value, indices=None):
setattr(self.venv, attr_name, value)
def render(self, mode="rgb_array"):
return self.venv.render()
def seed(self, seed=None):
if seed is not None:
self.venv._sim.seed(seed)
NUM_ENVS = 16
config = {
"policy_type": "MlpPolicy",
"total_timesteps": 1e10,
"env_name": "Rustoracer-v0",
}
run = wandb.init(
project="sb3",
config=config,
sync_tensorboard=True, # auto-upload sb3's tensorboard metrics
monitor_gym=True, # auto-upload the videos of agents playing the game
)
env = SB3VecAdapter(
RustoracerEnv(yaml="maps/berlin.yaml", num_envs=NUM_ENVS, render_mode="rgb_array")
)
env = VecMonitor(env)
class GCVecVideoRecorder(VecVideoRecorder):
def _stop_recording(self) -> None:
super()._stop_recording()
gc.collect()
env = GCVecVideoRecorder(
env,
f"videos/{run.id}",
record_video_trigger=lambda x: x % 10_000 == 0,
video_length=1_000,
)
model = PPO(config["policy_type"], env, verbose=1, tensorboard_log=f"runs/{run.id}")
model.learn(
total_timesteps=config["total_timesteps"],
callback=WandbCallback(
gradient_save_freq=100,
model_save_path=f"models/{run.id}",
verbose=2,
),
)
run.finish()