Repository for ECE Senior Design code

All of the code, including dependencies, are in catkin_ws. The autonomy assumes ROS Kinetic with Ubuntu 16.04; however, the system builds and runs as is in 18.04 with ROS Melodic.

To install the packages, run scripts/install.sh to pull all dependencies and build the workspace. Additionally, to run the stack, you need to install the DJI ROS Onboard SDK and generate an app key and app id with your DJI account. When running the SDK with the drone, you need to tell the drone that it is okay that the USB is still connected. This is done by running scripts/enable_usb.sh.

Inside of the workspace, the two main packages are mbz2020_common and mbz2020_planner. Common has the launch files, random scripts, and ROS message file definitions needed to get the whole system working. Inside the planner package are some dependencies, mbz2020_target_planners, and robot_motions_server_ros.

robot_motions_server_ros is the framework that wraps ROS ActionLib and allows for generic tasks. Inside mbz2020_target_planners are nodes for trajectory generation and the task config, where the pickup task/action is defined (i.e. mbz2020_target_planners/src/task_config/tasks/pickup_target.py).

The tools directory was used to label datasets, change path names, and convert file types.

Tools - Video labeling tool was a gui utilized to manually draw bounding boxes for labeling dataset. Label_directory_switch was used to change the path of each xml label. Train_test_switcharoni is used to randomize the data which is used to train vs the data used to test the neural network. Xml to csv was uzed to produce a file containing all the labelled data.

Object Detection - A package on github with slight modification. This is part of the tensorflow research package. The neural network uses this to train and run the neural network's inference graph.

object_detection_with_own_model - The code needed to run a visual test of the neural network. This also includes the localization algorithm.

The neural network inference graphs are in the google drive linked with this project.

Inside of scripts is the demo program we used to demonstrate the end effector working, including being able to command a pickup, detect a pickup, and command a drop and stop.