Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/PX4-Gazebo-Simulation/drl_uav
Deep reinforcement learning for UAV in Gazebo simulation environment
https://github.com/PX4-Gazebo-Simulation/drl_uav
Last synced: 21 days ago
JSON representation
Deep reinforcement learning for UAV in Gazebo simulation environment
- Host: GitHub
- URL: https://github.com/PX4-Gazebo-Simulation/drl_uav
- Owner: PX4-Gazebo-Simulation
- Created: 2018-07-09T05:06:07.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2018-08-02T15:43:20.000Z (over 6 years ago)
- Last Synced: 2024-08-05T17:24:07.496Z (4 months ago)
- Language: Python
- Homepage:
- Size: 53.7 KB
- Stars: 105
- Watchers: 4
- Forks: 29
- Open Issues: 1
-
Metadata Files:
- Readme: ReadMe.md
Awesome Lists containing this project
- awesome-Robot-Operating-Syetem - Deep reinforcement learning for UAV in Gazebo simulation environment
README
# Introduction
Deep reinforcement learning for UAV in Gazebo simulation environment
**Youtube:**
maunal control: https://www.youtube.com/watch?v=9zLjYLtHdPQ
training: https://www.youtube.com/watch?v=zbejm5uHPt8
## Environment:
Gazebo & pixhawk & ROS SITL(software in the loop) simulation:
## DRL:
+ state = [Pos_z, Vel_z, Thrust]
+ action = {0, 1, -1}
// 0: decrease thrust;
// 1: increase thrust;
// -1: environment needs to be restarted(manually selected)!
+ reward:
if(19.7 < Pos_z < 20.3) reward = 1
else reward = 0
+ Deep Network: 3 full connection layers
## Destination:
UAV hovering at the altitude of 20m.
# Requirements:
[Pixhawk & Gazebo](https://dev.px4.io/en/setup/dev_env_linux.html#gazebo-with-ros)
[ROS](http://wiki.ros.org/kinetic/Installation/Ubuntu)
[Tensorflow](https://www.tensorflow.org/install/install_linux)
[keras](https://keras.io/#installation)
# How to build the project
```
cd $HOME
mkdir src
```
```
cd ~/src
git clone https://github.com/PX4-Gazebo-Simulation/Firmware.git
cd Firmware
make px4fmu-v4_default
```
```
cd ~/src
mkdir -p mavros_ws/src
cd mavros_ws
catkin_init_workspace
cd src
git clone -b uavcomp https://github.com/PX4-Gazebo-Simulation/mavros.git
git clone -b uavcomp https://github.com/PX4-Gazebo-Simulation/mavlink
cd ..
catkin build
```
```
cd ~/src
mkdir -p attitude_controller/src
cd attitude_controller
catkin_init_workspace
cd src
git clone -b flight_test https://github.com/PX4-Gazebo-Simulation/state_machine.git
cd ..
catkin build
```
```
cd ~/src
mkdir -p DRL_node_ROS/src
cd DRL_node_ROS
catkin_init_workspace
cd src
git clone https://github.com/PX4-Gazebo-Simulation/drl_uav.git
cd ..
catkin build
```
# How to run UAV_DRL in Gazebo environment
(talker.py)
## 1. run pixhawk connection(MAVROS)
```
source ~/src/mavros_ws/devel/setup.bash
roslaunch mavros px4.launch fcu_url:="udp://:[email protected]:14557"
```
## 2. run pixhawk&gazebo
```
cd ~/src/Firmware
make posix_sitl_default gazebo
```
## 3. run state_machine: in branch flight_test
```
source ~/src/attitude_controller/devel/setup.bash
rosrun state_machine offb_simulation_test
```
## 4. switch pixhawk to offboard mode
```
source ~/src/mavros_ws/devel/setup.bash
rosrun mavros mavsafety arm
rosrun mavros mavsys mode -c OFFBOARD
```
## 5. run DRL
```
source ~/src/DRL_node_ROS/devel/setup.bash
rosrun drl_uav talker.py
```
# Constraints
## 1. Gazebo environment
Thrust: [0.40, 0.78]
Vel_z: [-3.0, 3.0]
Pos_z: [10, 30](for training; restart the system if current altitude is out of range)
## 2. UAV_DRL
if vel_z > 3.0 => force action=0(increase thrust)
if vel_z < -3.0 => force action=1(decrease thrust)
# Others
## 1. time delay of restart between pixhawk and DRL
~1.14s