-
Notifications
You must be signed in to change notification settings - Fork 0
/
eval_town2.py
182 lines (154 loc) · 7.08 KB
/
eval_town2.py
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
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
import queue
import carla
import numpy as np
import pandas as pd
import torch as th
from PIL import Image, ImageDraw
import pathlib
import tqdm
from rl_birdview.models.ppo_policy import PpoPolicy
from rl_train import get_obs_configs, get_env_wrapper_configs
from carla_gym.envs import LeaderboardEnv
from rl_birdview.utils.rl_birdview_wrapper import RlBirdviewWrapper
import carla_gym.utils.transforms as trans_utils
import gym
reward_configs = {
'hero': {
'entry_point': 'reward.valeo_action:ValeoAction',
'kwargs': {}
}
}
terminal_configs = {
'hero': {
'entry_point': 'terminal.valeo_no_det_px:ValeoNoDetPx',
'kwargs': {}
}
}
env_configs = {
'carla_map': 'Town02',
'weather_group': 'simple',
'routes_group': 'train'
}
def evaluate_agent(ckpt_path):
obs_configs = get_obs_configs(rgb=True)
env_wrapper_configs = get_env_wrapper_configs(rgb=True)
ckpt_key = ckpt_path.stem
eval_file_dir = pathlib.Path('agent_eval') / ckpt_key
eval_file_dir.mkdir(parents=True, exist_ok=True)
env = LeaderboardEnv(obs_configs=obs_configs, reward_configs=reward_configs,
terminal_configs=terminal_configs, host='localhost', port=2000,
seed=2021, no_rendering=False, **env_configs)
saved_variables = th.load(ckpt_path, map_location='cuda')
observation_space = {}
observation_space['birdview'] = gym.spaces.Box(low=0, high=255, shape=(3, 192, 192), dtype=np.uint8)
rgb_channels = 3
rgb_space = gym.spaces.Box(low=0, high=255, shape=(rgb_channels, 144, 256), dtype=np.uint8)
observation_space['central_rgb'] = rgb_space
observation_space['left_rgb'] = rgb_space
observation_space['right_rgb'] = rgb_space
observation_space['state'] = gym.spaces.Box(low=-10.0, high=30.0, shape=(6,), dtype=np.float32)
observation_space['traj'] = gym.spaces.Box(low=-10.0, high=30.0, shape=(10,), dtype=np.float32)
observation_space = gym.spaces.Dict(**observation_space)
policy_kwargs = {
'observation_space': observation_space,
'action_space': gym.spaces.Box(low=np.array([0, -1]), high=np.array([1, 1]), dtype=np.float32),
'policy_head_arch': [256, 256],
'value_head_arch': [256, 256],
'features_extractor_entry_point': 'rl_birdview.models.torch_layers:XtMaCNN',
'features_extractor_kwargs': {'states_neurons': [256,256]},
'distribution_entry_point': 'rl_birdview.models.distributions:BetaDistribution',
'fake_birdview': True
}
policy = PpoPolicy(**policy_kwargs)
policy.load_state_dict(saved_variables['policy_state_dict'])
device = 'cuda'
policy.to(device)
policy = policy.eval()
routes_completed = []
for route_id in tqdm.tqdm(range(6)):
episode_dir = eval_file_dir / ('route_%02d' % route_id)
(episode_dir / 'birdview').mkdir(parents=True)
(episode_dir / 'fake_birdview').mkdir(parents=True)
(episode_dir / 'central_rgb').mkdir(parents=True)
(episode_dir / 'left_rgb').mkdir(parents=True)
(episode_dir / 'right_rgb').mkdir(parents=True)
(episode_dir / 'traj_plot').mkdir(parents=True)
env.set_task_idx(route_id)
i_step = 0
x_pos_list = []
y_pos_list = []
z_pos_list = []
yaw_list = []
c_route_list = []
done_list = []
yaw_list = []
route_completed_list = []
raw_obs = env.reset()
ego_vehicle = env._ev_handler.ego_vehicles['hero']
global_route = ego_vehicle._global_route
c_route = False
done_ep = False
while not c_route and not done_ep:
obs = raw_obs['hero']
obs = RlBirdviewWrapper.process_obs(obs, env_wrapper_configs['input_states'])
obs_dict = dict([(k, np.stack([obs[k]])) for k in obs])
actions, _, _, _, _, _, fake_birdview = policy.forward(obs_dict, deterministic=True)
acc, steer = actions[0].astype(np.float64)
control = carla.VehicleControl(throttle=acc, steer=steer)
driver_control = {'hero': control}
birdview = obs['birdview']
for i_mask in range(1):
birdview_mask = birdview[i_mask * 3: i_mask * 3 + 3]
birdview_mask = np.transpose(birdview_mask, [1, 2, 0]).astype(np.uint8)
Image.fromarray(birdview_mask).save(episode_dir / 'birdview' / '{:0>4d}_{:0>2d}.png'.format(i_step, i_mask))
for i_mask in range(1):
birdview_mask = fake_birdview[0][i_mask * 3: i_mask * 3 + 3]
birdview_mask = np.transpose(birdview_mask, [1, 2, 0]).astype(np.uint8)
Image.fromarray(birdview_mask).save(episode_dir / 'fake_birdview' / '{:0>4d}_{:0>2d}.png'.format(i_step, i_mask))
central_rgb = obs['central_rgb']
central_rgb = np.transpose(central_rgb, [1, 2, 0]).astype(np.uint8)
Image.fromarray(central_rgb).save(episode_dir / 'central_rgb' / '{:0>4d}.png'.format(i_step))
left_rgb = obs['left_rgb']
left_rgb = np.transpose(left_rgb, [1, 2, 0]).astype(np.uint8)
Image.fromarray(left_rgb).save(episode_dir / 'left_rgb' / '{:0>4d}.png'.format(i_step))
right_rgb = obs['right_rgb']
right_rgb = np.transpose(right_rgb, [1, 2, 0]).astype(np.uint8)
Image.fromarray(right_rgb).save(episode_dir / 'right_rgb' / '{:0>4d}.png'.format(i_step))
traj_plot = obs['traj_plot'].numpy()
traj_plot = np.transpose(traj_plot, [1, 2, 0]).astype(np.uint8)
zero_img = np.zeros([traj_plot.shape[0], traj_plot.shape[0], 3], dtype=np.uint8)
zero_img[:,:,0:1] = traj_plot
Image.fromarray(zero_img).save(episode_dir / 'traj_plot' / '{:0>4d}.png'.format(i_step))
ev_transform = ego_vehicle.vehicle.get_transform()
ev_loc = ev_transform.location
ev_rot = ev_transform.rotation
x_pos_list.append(ev_loc.x)
y_pos_list.append(ev_loc.y)
z_pos_list.append(ev_loc.z)
yaw_list.append(ev_rot.yaw)
raw_obs, _, done, info = env.step(driver_control)
c_route = info['hero']['route_completion']['is_route_completed']
route_completed_in_m = info['hero']['route_completion']['route_completed_in_m']
route_completed_list.append(route_completed_in_m)
i_step += 1
done_ep = done['hero']
c_route_list.append(c_route)
done_list.append(done_ep)
ep_df = pd.DataFrame({
'x': x_pos_list,
'y': y_pos_list,
'z': z_pos_list,
'yaw': yaw_list,
'c_route': c_route_list,
'done': done_list,
'route_completed_in_m': route_completed_list
})
ep_df.to_json(episode_dir / 'episode.json')
routes_completed.append(c_route)
ep_df = pd.DataFrame({
'routes_completed': routes_completed
})
ep_df.to_json(eval_file_dir / 'eval.json')
if __name__ == '__main__':
ckpt_file = pathlib.Path('ckpt/ckpt_811008.pth')
evaluate_agent(ckpt_file)