YYDS!Python實(shí)現(xiàn)自動(dòng)駕駛
作者 | Veronica1312
來(lái)源丨CSDN博客
一、安裝環(huán)境gym是用于開發(fā)和比較強(qiáng)化學(xué)習(xí)算法的工具包,在python中安裝gym庫(kù)和其中子場(chǎng)景都較為簡(jiǎn)便。
安裝gym:
pip install gym
安裝自動(dòng)駕駛模塊,這里使用Edouard Leurent發(fā)布在github上的包highway-env(鏈接:https://github.com/eleurent/highway-env):
pip install --user git+https://github.com/eleurent/highway-env
其中包含6個(gè)場(chǎng)景:
高速公路——“highway-v0”
匯入——“merge-v0”
環(huán)島——“roundabout-v0”
泊車——“parking-v0”
十字路口——“intersection-v0”
賽車道——“racetrack-v0”
詳細(xì)文檔可以參考這里:
https://highway-env.readthedocs.io/en/latest/
二、配置環(huán)境安裝好后即可在代碼中進(jìn)行實(shí)驗(yàn)(以高速公路場(chǎng)景為例):
import gym
import highway_env
%matplotlib inline
env = gym.make('highway-v0')
env.reset()
for _ in range(3):
action = env.action_type.actions_indexes["IDLE"]
obs, reward, done, info = env.step(action)
env.render()
運(yùn)行后會(huì)在模擬器中生成如下場(chǎng)景:
綠色為ego vehicle env類有很多參數(shù)可以配置,具體可以參考原文檔。
三、訓(xùn)練模型1、數(shù)據(jù)處理(1)statehighway-env包中沒有定義傳感器,車輛所有的state (observations) 都從底層代碼讀取,節(jié)省了許多前期的工作量。根據(jù)文檔介紹,state (ovservations) 有三種輸出方式:Kinematics,Grayscale Image和Occupancy grid。
Kinematics
輸出V*F的矩陣,V代表需要觀測(cè)的車輛數(shù)量(包括ego vehicle本身),F(xiàn)代表需要統(tǒng)計(jì)的特征數(shù)量。例:
數(shù)據(jù)生成時(shí)會(huì)默認(rèn)歸一化,取值范圍:[100, 100, 20, 20],也可以設(shè)置ego vehicle以外的車輛屬性是地圖的絕對(duì)坐標(biāo)還是對(duì)ego vehicle的相對(duì)坐標(biāo)。
在定義環(huán)境時(shí)需要對(duì)特征的參數(shù)進(jìn)行設(shè)定:
config = \
{
"observation":
{
"type": "Kinematics",
#選取5輛車進(jìn)行觀察(包括ego vehicle)
"vehicles_count": 5,
#共7個(gè)特征
"features": ["presence", "x", "y", "vx", "vy", "cos_h", "sin_h"],
"features_range":
{
"x": [-100, 100],
"y": [-100, 100],
"vx": [-20, 20],
"vy": [-20, 20]
},
"absolute": False,
"order": "sorted"
},
"simulation_frequency": 8, # [Hz]
"policy_frequency": 2, # [Hz]
}
Grayscale Image
生成一張W*H的灰度圖像,W代表圖像寬度,H代表圖像高度
Occupancy grid
生成一個(gè)WHF的三維矩陣,用W*H的表格表示ego vehicle周圍的車輛情況,每個(gè)格子包含F(xiàn)個(gè)特征。
(2) actionhighway-env包中的action分為連續(xù)和離散兩種。連續(xù)型action可以直接定義throttle和steering angle的值,離散型包含5個(gè)meta actions:
(3) rewardACTIONS_ALL = {
0: 'LANE_LEFT',
1: 'IDLE',
2: 'LANE_RIGHT',
3: 'FASTER',
4: 'SLOWER'
}
highway-env包中除了泊車場(chǎng)景外都采用同一個(gè)reward function:
這個(gè)function只能在其源碼中更改,在外層只能調(diào)整權(quán)重。(泊車場(chǎng)景的reward function原文檔里有,懶得打公式了……)
2、搭建模型DQN網(wǎng)絡(luò)的結(jié)構(gòu)和搭建過程已經(jīng)在我另一篇文章中討論過,所以這里不再詳細(xì)解釋。我采用第一種state表示方式——Kinematics進(jìn)行示范。
由于state數(shù)據(jù)量較小(5輛車*7個(gè)特征),可以不考慮使用CNN,直接把二維數(shù)據(jù)的size[5,7]轉(zhuǎn)成[1,35]即可,模型的輸入就是35,輸出是離散action數(shù)量,共5個(gè)。
3、運(yùn)行結(jié)果import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.nn.functional as F
import torch.optim as optim
import torchvision.transforms as T
from torch import FloatTensor, LongTensor, ByteTensor
from collections import namedtuple
import random
Tensor = FloatTensor
EPSILON = 0 # epsilon used for epsilon greedy approach
GAMMA = 0.9
TARGET_NETWORK_REPLACE_FREQ = 40 # How frequently target netowrk updates
MEMORY_CAPACITY = 100
BATCH_SIZE = 80
LR = 0.01 # learning rate
class DQNNet(nn.Module):
def __init__(self):
super(DQNNet,self).__init__()
self.linear1 = nn.Linear(35,35)
self.linear2 = nn.Linear(35,5)
def forward(self,s):
s=torch.FloatTensor(s)
s = s.view(s.size(0),1,35)
s = self.linear1(s)
s = self.linear2(s)
return s
class DQN(object):
def __init__(self):
self.net,self.target_net = DQNNet(),DQNNet()
self.learn_step_counter = 0
self.memory = []
self.position = 0
self.capacity = MEMORY_CAPACITY
self.optimizer = torch.optim.Adam(self.net.parameters(), lr=LR)
self.loss_func = nn.MSELoss()
def choose_action(self,s,e):
x=np.expand_dims(s, axis=0)
if np.random.uniform() < 1-e:
actions_value = self.net.forward(x)
action = torch.max(actions_value,-1)[1].data.numpy()
action = action.max()
else:
action = np.random.randint(0, 5)
return action
def push_memory(self, s, a, r, s_):
if len(self.memory) < self.capacity:
self.memory.append(None)
self.memory[self.position] = Transition(torch.unsqueeze(torch.FloatTensor(s), 0),torch.unsqueeze(torch.FloatTensor(s_), 0),\
torch.from_numpy(np.array([a])),torch.from_numpy(np.array([r],dtype='float32')))#
self.position = (self.position + 1) % self.capacity
def get_sample(self,batch_size):
sample = random.sample(self.memory,batch_size)
return sample
def learn(self):
if self.learn_step_counter % TARGET_NETWORK_REPLACE_FREQ == 0:
self.target_net.load_state_dict(self.net.state_dict())
self.learn_step_counter += 1
transitions = self.get_sample(BATCH_SIZE)
batch = Transition(*zip(*transitions))
b_s = Variable(torch.cat(batch.state))
b_s_ = Variable(torch.cat(batch.next_state))
b_a = Variable(torch.cat(batch.action))
b_r = Variable(torch.cat(batch.reward))
q_eval = self.net.forward(b_s).squeeze(1).gather(1,b_a.unsqueeze(1).to(torch.int64))
q_next = self.target_net.forward(b_s_).detach() #
q_target = b_r + GAMMA * q_next.squeeze(1).max(1)[0].view(BATCH_SIZE, 1).t()
loss = self.loss_func(q_eval, q_target.t())
self.optimizer.zero_grad() # reset the gradient to zero
loss.backward()
self.optimizer.step() # execute back propagation for one step
return loss
Transition = namedtuple('Transition',('state', 'next_state','action', 'reward'))
各個(gè)部分都完成之后就可以組合在一起訓(xùn)練模型了,流程和用CARLA差不多,就不細(xì)說了。
初始化環(huán)境(DQN的類加進(jìn)去就行了):
import gym
import highway_env
from matplotlib import pyplot as plt
import numpy as np
import time
config = \
{
"observation":
{
"type": "Kinematics",
"vehicles_count": 5,
"features": ["presence", "x", "y", "vx", "vy", "cos_h", "sin_h"],
"features_range":
{
"x": [-100, 100],
"y": [-100, 100],
"vx": [-20, 20],
"vy": [-20, 20]
},
"absolute": False,
"order": "sorted"
},
"simulation_frequency": 8, # [Hz]
"policy_frequency": 2, # [Hz]
}
env = gym.make("highway-v0")
env.configure(config)
訓(xùn)練模型:
dqn=DQN()
count=0
reward=[]
avg_reward=0
all_reward=[]
time_=[]
all_time=[]
collision_his=[]
all_collision=[]
while True:
done = False
start_time=time.time()
s = env.reset()
while not done:
e = np.exp(-count/300) #隨機(jī)選擇action的概率,隨著訓(xùn)練次數(shù)增多逐漸降低
a = dqn.choose_action(s,e)
s_, r, done, info = env.step(a)
env.render()
dqn.push_memory(s, a, r, s_)
if ((dqn.position !=0)&(dqn.position % 99==0)):
loss_=dqn.learn()
count+=1
print('trained times:',count)
if (count%40==0):
avg_reward=np.mean(reward)
avg_time=np.mean(time_)
collision_rate=np.mean(collision_his)
all_reward.append(avg_reward)
all_time.append(avg_time)
all_collision.append(collision_rate)
plt.plot(all_reward)
plt.show()
plt.plot(all_time)
plt.show()
plt.plot(all_collision)
plt.show()
reward=[]
time_=[]
collision_his=[]
s = s_
reward.append(r)
end_time=time.time()
episode_time=end_time-start_time
time_.append(episode_time)
is_collision=1 if info['crashed']==True else 0
collision_his.append(is_collision)
我在代碼中添加了一些畫圖的函數(shù),在運(yùn)行過程中就可以掌握一些關(guān)鍵的指標(biāo),每訓(xùn)練40次統(tǒng)計(jì)一次平均值。
平均碰撞發(fā)生率:
epoch平均時(shí)長(zhǎng)(s):
平均reward:
可以看出平均碰撞發(fā)生率會(huì)隨訓(xùn)練次數(shù)增多逐漸降低,每個(gè)epoch持續(xù)的時(shí)間會(huì)逐漸延長(zhǎng)(如果發(fā)生碰撞epoch會(huì)立刻結(jié)束)
四、總結(jié)相比于我在之前文章中使用過的模擬器CARLA,highway-env環(huán)境包明顯更加抽象化,用類似游戲的表示方式,使得算法可以在一個(gè)理想的虛擬環(huán)境中得到訓(xùn)練,而不用考慮數(shù)據(jù)獲取方式、傳感器精度、運(yùn)算時(shí)長(zhǎng)等現(xiàn)實(shí)問題。對(duì)于端到端的算法設(shè)計(jì)和測(cè)試非常友好,但從自動(dòng)控制的角度來(lái)看,可以入手的方面較少,研究起來(lái)不太靈活。
*博客內(nèi)容為網(wǎng)友個(gè)人發(fā)布,僅代表博主個(gè)人觀點(diǎn),如有侵權(quán)請(qǐng)聯(lián)系工作人員刪除。