WSL_Ubuntu-20.04运行turtlebot3_mechaine_learning功能包跑DQN算法历程

一、系统信息

Win10 22H2
WSL2_Ubuntu-20.04
ROS-Noetic
Anaconda3

二、环境配置

1.安装CUDA驱动

在该网址下载安装英伟达驱动，安装时注意关闭杀毒软件
进入WSL输入 nvidia-smi查看显卡及驱动信息，若不报错并能成功显示则说明CUDA驱动已经在WSL配置成功

2.配置虚拟环境及软件包

用Anaconda创建一个python版本为3.6的虚拟环境

conda create -n tf python=3.6

激活虚拟环境

conda activate tf

安装cuda、cudnn

#30系以下显卡
conda install cudatoolkit==9.0
conda install cudnn==7.1.2

#30系以上显卡
conda install cudatoolkit==11.2.0
conda install cudnn==8.1.0.77

指定清华源安装ROS依赖包

pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple --upgrade -U rosinstall msgpack empy defusedxml netifaces

安装tensorflow-gpu

#30系以下显卡
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple --upgrade tensorflow-gpu==1.8

#30系以上显卡
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple --upgrade tensorflow-gpu==2.5.0

安装keras

#30系以下显卡
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple --upgrade keras==2.1.6

#30系以上显卡
pip3 install keras-nightly~=2.5.0.dev --trusted-host pypi.org --trusted-host files.pythonhosted.org

安装empy

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple --upgrade empy==3.3.4

安装画图工具包

pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple --upgrade pyqtgraph
conda install pyqt

降低numpy版本

#30系以下显卡
pip install -U numpy==1.14.5 -i https://pypi.tuna.tsinghua.edu.cn/simple

3.检测环境是否配置成功

在WSL中输入

python
import tensorflow as tf
print(tf.test.is_gpu_available())
import keras

若结果是True且无警告或报错则表明tensorflow配置成功

三、拷贝源码及修改配置

1.建立工作空间

mkdir -p catkin_ws/src
cd ~/catkin_ws/src

2.拷贝源码

git clone https://gitee.com/zthcool/turtlebot3_msgs.git
git clone https://gitee.com/zthcool/turtlebot3.git
git clone https://gitee.com/zthcool/turtlebot3_simulations.git
git clone https://gitee.com/oliver_zxh/turtlebot3_machine_learning.git

3.编译及配置环境变量

cd ~/catkin_ws && catkin_make
sudo vim ~/.bashrc
#在文件末尾添加以下内容
source /home/xxx/catkin_ws/devel/setup.bash
export TURTLEBOT3_MODEL=waffle

#刷新环境变量
source ~/.bashrc

4.设置参数及修正bug

(1) 打开源码文件turtlebot3/turtlebot3_description/urdf/turtlebot3_waffle.gazebo.xacro

<xacro:arg name="laser_visual" default="false"/>   # 如果想看到激光扫描线,设置成 'true'

<scan>
  <horizontal>
    <samples>360</samples>            # 修改成24
    <resolution>1</resolution>
    <min_angle>0.0</min_angle>
    <max_angle>6.28319</max_angle>
  </horizontal>
</scan>

(2) 打开源码文件turtlebot3_simulations/turtlebot3_gazebo/launch/turtlebot3_stage_1.launch
将

<param name="robot_description" command="$(find xacro)/xacro.py '$(find carsim_discription)/urdf/model.urdf'" />

修改为

<param name="robot_description" command="$(find xacro)/xacro '$(find carsim_discription)/urdf/model.urdf'" />

(3) 打开源码文件turtlebot3_machine_learning/turtlebot3_dqn/src/turtlebot3_dqn/environment_stage_1.py
将

from respawnGoal import Respawn

修改为

from src.turtlebot3_dqn.respawnGoal import Respawn

(4) 打开源码文件turtlebot3_machine_learning/turtlebot3_dqn/nodes/turtlebot3_dqn_stage_1
将

from keras.models import Sequential, load_model
from keras.optimizers import RMSprop
from keras.layers import Dense, Dropout, Activation

修改为

from tensorflow.keras.models import Sequential, load_model
from tensorflow.keras.optimizers import RMSprop
from tensorflow.keras.layers import Dense, Dropout, Activation

再次编译工作空间

cd ~/catkin_ws  && catkin_make

四、运行程序

打开第一个终端，开启仿真环境

conda activate tf
roslaunch turtlebot3_gazebo turtlebot3_stage_1.launch

打开第二个终端，运行DQN算法

conda activate tf
 
roslaunch turtlebot3_dqn turtlebot3_dqn_stage_1.launch

打开第三个终端，启动数据图形显示节点

conda activate tf
roslaunch turtlebot3_dqn result_graph.launch

五、仿真结果

gazebo环境可正常打开，但提示一个警告

Node::Advertise(): Error advertising service [/goal/shininess]. Did you forget to start the discovery service?

DQN算法程序可正常加载
训练数小时后，计算图显示结果如图所示：

下一步计划研究提示该警告的原因，并尽量去除警告。