Implemented the vision and control on a tiny drone and statisfied the following milestones
- You can control your drone using some form of ROS based "keyboard" control that sends commands to the drone for how it should move.
- You can collect data into a rosbag and play back as a minimum pose and camera images
- Your drone can fly to set points. If the drone starts from the floor you should be able to fly roughly to a point specified by some in the teaching team without having to recompile the program.
- You can perform an experiment where the drone starts from the floor, takes off, flies Xm forward, rotates (N*360+180) degrees and flies Xm forward. You should be able to present how the difference in position and orientation at the end depends on X and N roughly. In a perfect world it should come back to exactly the same position.
- Using the feed from the camera of a flying drone you can detect at least one object. You are free to pick the object and you can decide how to fly (even manually with functionality from item 1) but you must mark where in the image the object is detected. No object detected should mean nothing marked in the image so that one can say exact when it is detected and not.
Everything that is mentioned in milestones 1 is assumed here as well.
- The drone can position itself using a marker. This can be demonstrated by moving the marker and the drone should follow to keep the relative pose constant.
- We want you to be able to attempt the clear checkpoint routine (i.e. stand still, rotate 360, stand still) but we do not require you to recover the position after the rotation.
- You can plan a path from a current/given point to a checkpoint. 2D path planning is sufficient, i.e. if 3D motion is needed to find a path it is OK if you fail. You do not need to be able to execute the path but it should be plausible that the path could be followed with small modifications.
- For at least one object you can detect it and position it so that you can visualise it in rviz such that when the position of the drone is correct it overlaps with the map position.
- You can perform this detection and positioning of objects when the drone is flying. Looking at the detected objects and comparing this to the true position of the object in the map should thus tell you something about the error in your drone's position and you should be able to discuss that.
Everything that is mentioned in milestones 1 and 2 is assumed here as well.
Only things that are integrated into one system counts here. If there are several pieces that cannot run together it is the smallest integrated piece that counts as your integrated system.
- Starting from an unknown position (ArUco marker visible at take-off), your integrated system can plan+navigate automatically and clear at least one non-ArUco checkpoint. That is the drone has to fly around something to get there.