The vision-based-avoidance environment is being developed by Caleb Harris at Georgia Institute of Technology. For more information please contact the developer.
Follow these instructions to install the project from source. More information will be provided in the future, but for now see the instructions for RotorS at my branch here. Another good source for installing the proper setup for computer vision and deep learning tools is found here.
This package is dependent on RotorS, mav_control_rw, pcl, cv2, torch and more! For more detailed package requirements please see the requirements file and the ROS package file.
The key files and folders to consider are:
- *dnn\scripts*
- dnn_architecture.py
- train_dnn.py
- scripts/
- astar_data_collection.py
- path_planning.py
- dagger_main.py
- dnn_policy
- path_planner
- dagger
The original RotorS environment provided multiple MAV models such as hummingbird and firefly. Additional models have been created and are described below:
- DJI: A custom designed quadcopter which uses dji F450 flamewheel frame
- matrice: A model of the dji matrice M100M
The original RotorS environment and Control packages provided PID and MPC controllers for the vehicles. Personal updates has extended these controllers to the new MAVs, and has modified the MPC controller using the ARUCO solver to utilize rate-based model tracking and tune the parameters for the new MAVs.
Start by running simulation with one of the MAVs and worlds:
roslaunch rotors_gazebo obstacle_avoidance.launch mav_name:=DJI world_name:=ob
Then, run the path planning expert algorithm (path_planning.py)
Then run the following command:
rosbag record rosout tf /DJI/command/trajectory /DJI/command/motor_speed /DJI/command/roll_pitch_yawrate_thrust /DJI/ground_truth/odometry /DJI/odometry_sensor1/odometry /DJI/vi_sensor/camera_depth/camera/image_raw /DJI/vi_sensor/camera_depth/camera/image_raw/compressed /initialpose /move_base_simple/goal
Afterwards, the file "read_bag.py" will be used to translate the data into a dataset.
To collect data for training follow the process below.
First, launch the simulation with the desired UAV and world. The octomap server and MPC controller will also start:
roslaunch rotors_gazebo obstacle_avoidance.launch mav_name:=DJI world_name:=obstacle_two
Second, launch the Astar solution script that is for data collection: (Currently it is not available in rotors_gazebo)
rosrun rotors_gazebo astar_data_collection.py DJI
Next, launch the data collection node:
rosrun rotors_gazebo data_collection.py
Lastly, run the waypoint publisher file for data collection:
rosrun rotors_gazebo waypoint_publisher_file_new 0 0 1 -45 ~/test.txt __ns:=DJI
To run the current world model and controller run:
roslaunch rotors_gazebo obstacle_avoidance.launch
To run with joystick:
roslaunch rotors_gazebo obstacle_avoidance_with_joy.launch
To run using linear mpc controller:
roslaunch rotors_gazebo obstacle_avoidance_linear.launch
Note: Currently, the world and launch files are moved into the rotors_gazebo folders for easy development. In the future the launch file will be updated to access the files relative to this package.
To run the waypoint publisher python script:
rosrun dagger_pytorch_ros waypoint_publisher.py $CSV_FILE_LOCATION
The csv file is organized by having the MAV namespace on the first line, then any set of waypoints with x, y, z, yaw.
firefly
0, 0, 0.1, 0
0, 0, 1, 0
1, 0, 1, 0
1, 1, 1, 0
0, 0, 0.1, 0
To build a map via the Octomap package, first start the environment with any controller. Then, start the octomap server:
roslaunch octomap_server octomap_mapping.launch
Send the vehicle along a preset trajectory for mapping, ex. obstacle_trajectory.csv
After completing this trajectory, or checking the map is complete, save the map respectively as a binary map and probability map:
rosrun octomap_saver mafile.bt
octomap_saver -f mapfile.ot
Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests to us.
- Caleb Harris - Developer and Maintaner - (caleb.harris94@gatech.edu)
- "There is good in this world Frodo, and it is worth fighting for." - Sam Baggins, Lord of the Rings