doggystyle-star / uav_tracking Goto Github PK

After about 700 episodes, the performance is shown as follows:

Python 3.10
Tensorflow 2.14.0
tensorlayer 2.2.5

• gym_examples/: Custom Environment: The agent (blue dot) navigates through both static and dynamic obstacles (black) to reach the goal (green).
• ddpg_model/: The folder that stores the model weights in each episode.
• gym_examples/envs/gazebo_world.py: The detailed implementation of the env gazebo_world.
• gym_examples/envs/multirotor_communication.py: Main file1, Start the communication of UAV and the env gazebo_world.
• gym_examples/envs/tracking_IBVS.py: Main file2, the controller of the env gazebo_world.
• DDPG_UAV.py: Main file3， enhancing DDPG with Robust Controller for Accelerated Training.

"The DDPG algorithm is used for calcuating the parameter of robust controller, comeared to pure Reinforcement learning it is faster and using only cpu computing resources."

Modify some launch files

• cp -r gazeboworld/outdoor.launch* ~/PX4_Firmware/launch/
• cp -r gazeboworld/world_nightmare.world* ~/PX4_Firmware/Tools/sitl_gazebo/worlds/

• roslaunch px4 outdoor.launch # world

• roslaunch yolov5_ros yolov5.launch # find target

• python3 multirotor_communication.py iris 0 # for communication

• python3 multirotor_keboard_control.py iris 1 vel # for control uav to move

• Close the file multirotor_keboard_control.py

• python3 tracking_IBVS.py # for tracking

• python3 DDPG_UAV.py # for training

• python3 DDPG_UAV.py --mode test --save_path /path/to/your/model # for testing